1 /*M///////////////////////////////////////////////////////////////////////////////////////
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
5 // By downloading, copying, installing or using the software you agree to this license.
6 // If you do not agree to this license, do not download, install,
7 // copy or use the software.
11 // For Open Source Computer Vision Library
13 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
14 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
15 // Third party copyrights are property of their respective owners.
17 // Redistribution and use in source and binary forms, with or without modification,
18 // are permitted provided that the following conditions are met:
20 // * Redistribution's of source code must retain the above copyright notice,
21 // this list of conditions and the following disclaimer.
23 // * Redistribution's in binary form must reproduce the above copyright notice,
24 // this list of conditions and the following disclaimer in the documentation
25 // and/or other materials provided with the distribution.
27 // * The name of the copyright holders may not be used to endorse or promote products
28 // derived from this software without specific prior written permission.
30 // This software is provided by the copyright holders and contributors "as is" and
31 // any express or implied warranties, including, but not limited to, the implied
32 // warranties of merchantability and fitness for a particular purpose are disclaimed.
33 // In no event shall the Intel Corporation or contributors be liable for any direct,
34 // indirect, incidental, special, exemplary, or consequential damages
35 // (including, but not limited to, procurement of substitute goods or services;
36 // loss of use, data, or profits; or business interruption) however caused
37 // and on any theory of liability, whether in contract, strict liability,
38 // or tort (including negligence or otherwise) arising in any way out of
39 // the use of this software, even if advised of the possibility of such damage.
43 /* ////////////////////////////////////////////////////////////////////
45 // Geometrical transforms on images and matrices: rotation, zoom etc.
49 #include "precomp.hpp"
50 #include "opencl_kernels.hpp"
52 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
53 static IppStatus sts = ippInit();
58 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR*100 + IPP_VERSION_MINOR >= 701)
59 typedef IppStatus (CV_STDCALL* ippiResizeFunc)(const void*, int, const void*, int, IppiPoint, IppiSize, IppiBorderType, void*, void*, Ipp8u*);
62 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
63 typedef IppStatus (CV_STDCALL* ippiSetFunc)(const void*, void *, int, IppiSize);
64 typedef IppStatus (CV_STDCALL* ippiWarpPerspectiveBackFunc)(const void*, IppiSize, int, IppiRect, void *, int, IppiRect, double [3][3], int);
65 typedef IppStatus (CV_STDCALL* ippiWarpAffineBackFunc)(const void*, IppiSize, int, IppiRect, void *, int, IppiRect, double [2][3], int);
67 template <int channels, typename Type>
68 bool IPPSetSimple(cv::Scalar value, void *dataPointer, int step, IppiSize &size, ippiSetFunc func)
70 Type values[channels];
71 for( int i = 0; i < channels; i++ )
72 values[i] = (Type)value[i];
73 return func(values, dataPointer, step, size) >= 0;
76 bool IPPSet(const cv::Scalar &value, void *dataPointer, int step, IppiSize &size, int channels, int depth)
83 return ippiSet_8u_C1R((Ipp8u)value[0], (Ipp8u *)dataPointer, step, size) >= 0;
85 return ippiSet_16u_C1R((Ipp16u)value[0], (Ipp16u *)dataPointer, step, size) >= 0;
87 return ippiSet_32f_C1R((Ipp32f)value[0], (Ipp32f *)dataPointer, step, size) >= 0;
97 return IPPSetSimple<3, Ipp8u>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_8u_C3R);
99 return IPPSetSimple<3, Ipp16u>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_16u_C3R);
101 return IPPSetSimple<3, Ipp32f>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_32f_C3R);
104 else if( channels == 4 )
109 return IPPSetSimple<4, Ipp8u>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_8u_C4R);
111 return IPPSetSimple<4, Ipp16u>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_16u_C4R);
113 return IPPSetSimple<4, Ipp32f>(value, dataPointer, step, size, (ippiSetFunc)ippiSet_32f_C4R);
121 /************** interpolation formulas and tables ***************/
123 const int INTER_RESIZE_COEF_BITS=11;
124 const int INTER_RESIZE_COEF_SCALE=1 << INTER_RESIZE_COEF_BITS;
126 const int INTER_REMAP_COEF_BITS=15;
127 const int INTER_REMAP_COEF_SCALE=1 << INTER_REMAP_COEF_BITS;
129 static uchar NNDeltaTab_i[INTER_TAB_SIZE2][2];
131 static float BilinearTab_f[INTER_TAB_SIZE2][2][2];
132 static short BilinearTab_i[INTER_TAB_SIZE2][2][2];
135 static short BilinearTab_iC4_buf[INTER_TAB_SIZE2+2][2][8];
136 static short (*BilinearTab_iC4)[2][8] = (short (*)[2][8])alignPtr(BilinearTab_iC4_buf, 16);
139 static float BicubicTab_f[INTER_TAB_SIZE2][4][4];
140 static short BicubicTab_i[INTER_TAB_SIZE2][4][4];
142 static float Lanczos4Tab_f[INTER_TAB_SIZE2][8][8];
143 static short Lanczos4Tab_i[INTER_TAB_SIZE2][8][8];
145 static inline void interpolateLinear( float x, float* coeffs )
151 static inline void interpolateCubic( float x, float* coeffs )
153 const float A = -0.75f;
155 coeffs[0] = ((A*(x + 1) - 5*A)*(x + 1) + 8*A)*(x + 1) - 4*A;
156 coeffs[1] = ((A + 2)*x - (A + 3))*x*x + 1;
157 coeffs[2] = ((A + 2)*(1 - x) - (A + 3))*(1 - x)*(1 - x) + 1;
158 coeffs[3] = 1.f - coeffs[0] - coeffs[1] - coeffs[2];
161 static inline void interpolateLanczos4( float x, float* coeffs )
163 static const double s45 = 0.70710678118654752440084436210485;
164 static const double cs[][2]=
165 {{1, 0}, {-s45, -s45}, {0, 1}, {s45, -s45}, {-1, 0}, {s45, s45}, {0, -1}, {-s45, s45}};
167 if( x < FLT_EPSILON )
169 for( int i = 0; i < 8; i++ )
176 double y0=-(x+3)*CV_PI*0.25, s0 = sin(y0), c0=cos(y0);
177 for(int i = 0; i < 8; i++ )
179 double y = -(x+3-i)*CV_PI*0.25;
180 coeffs[i] = (float)((cs[i][0]*s0 + cs[i][1]*c0)/(y*y));
185 for(int i = 0; i < 8; i++ )
189 static void initInterTab1D(int method, float* tab, int tabsz)
191 float scale = 1.f/tabsz;
192 if( method == INTER_LINEAR )
194 for( int i = 0; i < tabsz; i++, tab += 2 )
195 interpolateLinear( i*scale, tab );
197 else if( method == INTER_CUBIC )
199 for( int i = 0; i < tabsz; i++, tab += 4 )
200 interpolateCubic( i*scale, tab );
202 else if( method == INTER_LANCZOS4 )
204 for( int i = 0; i < tabsz; i++, tab += 8 )
205 interpolateLanczos4( i*scale, tab );
208 CV_Error( CV_StsBadArg, "Unknown interpolation method" );
212 static const void* initInterTab2D( int method, bool fixpt )
214 static bool inittab[INTER_MAX+1] = {false};
218 if( method == INTER_LINEAR )
219 tab = BilinearTab_f[0][0], itab = BilinearTab_i[0][0], ksize=2;
220 else if( method == INTER_CUBIC )
221 tab = BicubicTab_f[0][0], itab = BicubicTab_i[0][0], ksize=4;
222 else if( method == INTER_LANCZOS4 )
223 tab = Lanczos4Tab_f[0][0], itab = Lanczos4Tab_i[0][0], ksize=8;
225 CV_Error( CV_StsBadArg, "Unknown/unsupported interpolation type" );
227 if( !inittab[method] )
229 AutoBuffer<float> _tab(8*INTER_TAB_SIZE);
231 initInterTab1D(method, _tab, INTER_TAB_SIZE);
232 for( i = 0; i < INTER_TAB_SIZE; i++ )
233 for( j = 0; j < INTER_TAB_SIZE; j++, tab += ksize*ksize, itab += ksize*ksize )
236 NNDeltaTab_i[i*INTER_TAB_SIZE+j][0] = j < INTER_TAB_SIZE/2;
237 NNDeltaTab_i[i*INTER_TAB_SIZE+j][1] = i < INTER_TAB_SIZE/2;
239 for( k1 = 0; k1 < ksize; k1++ )
241 float vy = _tab[i*ksize + k1];
242 for( k2 = 0; k2 < ksize; k2++ )
244 float v = vy*_tab[j*ksize + k2];
245 tab[k1*ksize + k2] = v;
246 isum += itab[k1*ksize + k2] = saturate_cast<short>(v*INTER_REMAP_COEF_SCALE);
250 if( isum != INTER_REMAP_COEF_SCALE )
252 int diff = isum - INTER_REMAP_COEF_SCALE;
253 int ksize2 = ksize/2, Mk1=ksize2, Mk2=ksize2, mk1=ksize2, mk2=ksize2;
254 for( k1 = ksize2; k1 < ksize2+2; k1++ )
255 for( k2 = ksize2; k2 < ksize2+2; k2++ )
257 if( itab[k1*ksize+k2] < itab[mk1*ksize+mk2] )
259 else if( itab[k1*ksize+k2] > itab[Mk1*ksize+Mk2] )
263 itab[Mk1*ksize + Mk2] = (short)(itab[Mk1*ksize + Mk2] - diff);
265 itab[mk1*ksize + mk2] = (short)(itab[mk1*ksize + mk2] - diff);
268 tab -= INTER_TAB_SIZE2*ksize*ksize;
269 itab -= INTER_TAB_SIZE2*ksize*ksize;
271 if( method == INTER_LINEAR )
273 for( i = 0; i < INTER_TAB_SIZE2; i++ )
274 for( j = 0; j < 4; j++ )
276 BilinearTab_iC4[i][0][j*2] = BilinearTab_i[i][0][0];
277 BilinearTab_iC4[i][0][j*2+1] = BilinearTab_i[i][0][1];
278 BilinearTab_iC4[i][1][j*2] = BilinearTab_i[i][1][0];
279 BilinearTab_iC4[i][1][j*2+1] = BilinearTab_i[i][1][1];
283 inittab[method] = true;
285 return fixpt ? (const void*)itab : (const void*)tab;
289 static bool initAllInterTab2D()
291 return initInterTab2D( INTER_LINEAR, false ) &&
292 initInterTab2D( INTER_LINEAR, true ) &&
293 initInterTab2D( INTER_CUBIC, false ) &&
294 initInterTab2D( INTER_CUBIC, true ) &&
295 initInterTab2D( INTER_LANCZOS4, false ) &&
296 initInterTab2D( INTER_LANCZOS4, true );
299 static volatile bool doInitAllInterTab2D = initAllInterTab2D();
302 template<typename ST, typename DT> struct Cast
307 DT operator()(ST val) const { return saturate_cast<DT>(val); }
310 template<typename ST, typename DT, int bits> struct FixedPtCast
314 enum { SHIFT = bits, DELTA = 1 << (bits-1) };
316 DT operator()(ST val) const { return saturate_cast<DT>((val + DELTA)>>SHIFT); }
319 /****************************************************************************************\
321 \****************************************************************************************/
323 class resizeNNInvoker :
324 public ParallelLoopBody
327 resizeNNInvoker(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
328 ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
333 virtual void operator() (const Range& range) const
335 Size ssize = src.size(), dsize = dst.size();
336 int y, x, pix_size = (int)src.elemSize();
338 for( y = range.start; y < range.end; y++ )
340 uchar* D = dst.data + dst.step*y;
341 int sy = std::min(cvFloor(y*ify), ssize.height-1);
342 const uchar* S = src.data + src.step*sy;
347 for( x = 0; x <= dsize.width - 2; x += 2 )
349 uchar t0 = S[x_ofs[x]];
350 uchar t1 = S[x_ofs[x+1]];
355 for( ; x < dsize.width; x++ )
359 for( x = 0; x < dsize.width; x++ )
360 *(ushort*)(D + x*2) = *(ushort*)(S + x_ofs[x]);
363 for( x = 0; x < dsize.width; x++, D += 3 )
365 const uchar* _tS = S + x_ofs[x];
366 D[0] = _tS[0]; D[1] = _tS[1]; D[2] = _tS[2];
370 for( x = 0; x < dsize.width; x++ )
371 *(int*)(D + x*4) = *(int*)(S + x_ofs[x]);
374 for( x = 0; x < dsize.width; x++, D += 6 )
376 const ushort* _tS = (const ushort*)(S + x_ofs[x]);
377 ushort* _tD = (ushort*)D;
378 _tD[0] = _tS[0]; _tD[1] = _tS[1]; _tD[2] = _tS[2];
382 for( x = 0; x < dsize.width; x++, D += 8 )
384 const int* _tS = (const int*)(S + x_ofs[x]);
386 _tD[0] = _tS[0]; _tD[1] = _tS[1];
390 for( x = 0; x < dsize.width; x++, D += 12 )
392 const int* _tS = (const int*)(S + x_ofs[x]);
394 _tD[0] = _tS[0]; _tD[1] = _tS[1]; _tD[2] = _tS[2];
398 for( x = 0; x < dsize.width; x++, D += pix_size )
400 const int* _tS = (const int*)(S + x_ofs[x]);
402 for( int k = 0; k < pix_size4; k++ )
412 int* x_ofs, pix_size4;
415 resizeNNInvoker(const resizeNNInvoker&);
416 resizeNNInvoker& operator=(const resizeNNInvoker&);
420 resizeNN( const Mat& src, Mat& dst, double fx, double fy )
422 Size ssize = src.size(), dsize = dst.size();
423 AutoBuffer<int> _x_ofs(dsize.width);
425 int pix_size = (int)src.elemSize();
426 int pix_size4 = (int)(pix_size / sizeof(int));
427 double ifx = 1./fx, ify = 1./fy;
430 for( x = 0; x < dsize.width; x++ )
432 int sx = cvFloor(x*ifx);
433 x_ofs[x] = std::min(sx, ssize.width-1)*pix_size;
436 Range range(0, dsize.height);
437 resizeNNInvoker invoker(src, dst, x_ofs, pix_size4, ify);
438 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
444 int operator()(const uchar**, uchar*, const uchar*, int ) const { return 0; }
449 int operator()(const uchar**, uchar**, int, const int*,
450 const uchar*, int, int, int, int, int) const { return 0; }
455 struct VResizeLinearVec_32s8u
457 int operator()(const uchar** _src, uchar* dst, const uchar* _beta, int width ) const
459 if( !checkHardwareSupport(CV_CPU_SSE2) )
462 const int** src = (const int**)_src;
463 const short* beta = (const short*)_beta;
464 const int *S0 = src[0], *S1 = src[1];
466 __m128i b0 = _mm_set1_epi16(beta[0]), b1 = _mm_set1_epi16(beta[1]);
467 __m128i delta = _mm_set1_epi16(2);
469 if( (((size_t)S0|(size_t)S1)&15) == 0 )
470 for( ; x <= width - 16; x += 16 )
472 __m128i x0, x1, x2, y0, y1, y2;
473 x0 = _mm_load_si128((const __m128i*)(S0 + x));
474 x1 = _mm_load_si128((const __m128i*)(S0 + x + 4));
475 y0 = _mm_load_si128((const __m128i*)(S1 + x));
476 y1 = _mm_load_si128((const __m128i*)(S1 + x + 4));
477 x0 = _mm_packs_epi32(_mm_srai_epi32(x0, 4), _mm_srai_epi32(x1, 4));
478 y0 = _mm_packs_epi32(_mm_srai_epi32(y0, 4), _mm_srai_epi32(y1, 4));
480 x1 = _mm_load_si128((const __m128i*)(S0 + x + 8));
481 x2 = _mm_load_si128((const __m128i*)(S0 + x + 12));
482 y1 = _mm_load_si128((const __m128i*)(S1 + x + 8));
483 y2 = _mm_load_si128((const __m128i*)(S1 + x + 12));
484 x1 = _mm_packs_epi32(_mm_srai_epi32(x1, 4), _mm_srai_epi32(x2, 4));
485 y1 = _mm_packs_epi32(_mm_srai_epi32(y1, 4), _mm_srai_epi32(y2, 4));
487 x0 = _mm_adds_epi16(_mm_mulhi_epi16( x0, b0 ), _mm_mulhi_epi16( y0, b1 ));
488 x1 = _mm_adds_epi16(_mm_mulhi_epi16( x1, b0 ), _mm_mulhi_epi16( y1, b1 ));
490 x0 = _mm_srai_epi16(_mm_adds_epi16(x0, delta), 2);
491 x1 = _mm_srai_epi16(_mm_adds_epi16(x1, delta), 2);
492 _mm_storeu_si128( (__m128i*)(dst + x), _mm_packus_epi16(x0, x1));
495 for( ; x <= width - 16; x += 16 )
497 __m128i x0, x1, x2, y0, y1, y2;
498 x0 = _mm_loadu_si128((const __m128i*)(S0 + x));
499 x1 = _mm_loadu_si128((const __m128i*)(S0 + x + 4));
500 y0 = _mm_loadu_si128((const __m128i*)(S1 + x));
501 y1 = _mm_loadu_si128((const __m128i*)(S1 + x + 4));
502 x0 = _mm_packs_epi32(_mm_srai_epi32(x0, 4), _mm_srai_epi32(x1, 4));
503 y0 = _mm_packs_epi32(_mm_srai_epi32(y0, 4), _mm_srai_epi32(y1, 4));
505 x1 = _mm_loadu_si128((const __m128i*)(S0 + x + 8));
506 x2 = _mm_loadu_si128((const __m128i*)(S0 + x + 12));
507 y1 = _mm_loadu_si128((const __m128i*)(S1 + x + 8));
508 y2 = _mm_loadu_si128((const __m128i*)(S1 + x + 12));
509 x1 = _mm_packs_epi32(_mm_srai_epi32(x1, 4), _mm_srai_epi32(x2, 4));
510 y1 = _mm_packs_epi32(_mm_srai_epi32(y1, 4), _mm_srai_epi32(y2, 4));
512 x0 = _mm_adds_epi16(_mm_mulhi_epi16( x0, b0 ), _mm_mulhi_epi16( y0, b1 ));
513 x1 = _mm_adds_epi16(_mm_mulhi_epi16( x1, b0 ), _mm_mulhi_epi16( y1, b1 ));
515 x0 = _mm_srai_epi16(_mm_adds_epi16(x0, delta), 2);
516 x1 = _mm_srai_epi16(_mm_adds_epi16(x1, delta), 2);
517 _mm_storeu_si128( (__m128i*)(dst + x), _mm_packus_epi16(x0, x1));
520 for( ; x < width - 4; x += 4 )
523 x0 = _mm_srai_epi32(_mm_loadu_si128((const __m128i*)(S0 + x)), 4);
524 y0 = _mm_srai_epi32(_mm_loadu_si128((const __m128i*)(S1 + x)), 4);
525 x0 = _mm_packs_epi32(x0, x0);
526 y0 = _mm_packs_epi32(y0, y0);
527 x0 = _mm_adds_epi16(_mm_mulhi_epi16(x0, b0), _mm_mulhi_epi16(y0, b1));
528 x0 = _mm_srai_epi16(_mm_adds_epi16(x0, delta), 2);
529 x0 = _mm_packus_epi16(x0, x0);
530 *(int*)(dst + x) = _mm_cvtsi128_si32(x0);
538 template<int shiftval> struct VResizeLinearVec_32f16
540 int operator()(const uchar** _src, uchar* _dst, const uchar* _beta, int width ) const
542 if( !checkHardwareSupport(CV_CPU_SSE2) )
545 const float** src = (const float**)_src;
546 const float* beta = (const float*)_beta;
547 const float *S0 = src[0], *S1 = src[1];
548 ushort* dst = (ushort*)_dst;
551 __m128 b0 = _mm_set1_ps(beta[0]), b1 = _mm_set1_ps(beta[1]);
552 __m128i preshift = _mm_set1_epi32(shiftval);
553 __m128i postshift = _mm_set1_epi16((short)shiftval);
555 if( (((size_t)S0|(size_t)S1)&15) == 0 )
556 for( ; x <= width - 16; x += 16 )
558 __m128 x0, x1, y0, y1;
560 x0 = _mm_load_ps(S0 + x);
561 x1 = _mm_load_ps(S0 + x + 4);
562 y0 = _mm_load_ps(S1 + x);
563 y1 = _mm_load_ps(S1 + x + 4);
565 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
566 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
567 t0 = _mm_add_epi32(_mm_cvtps_epi32(x0), preshift);
568 t2 = _mm_add_epi32(_mm_cvtps_epi32(x1), preshift);
569 t0 = _mm_add_epi16(_mm_packs_epi32(t0, t2), postshift);
571 x0 = _mm_load_ps(S0 + x + 8);
572 x1 = _mm_load_ps(S0 + x + 12);
573 y0 = _mm_load_ps(S1 + x + 8);
574 y1 = _mm_load_ps(S1 + x + 12);
576 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
577 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
578 t1 = _mm_add_epi32(_mm_cvtps_epi32(x0), preshift);
579 t2 = _mm_add_epi32(_mm_cvtps_epi32(x1), preshift);
580 t1 = _mm_add_epi16(_mm_packs_epi32(t1, t2), postshift);
582 _mm_storeu_si128( (__m128i*)(dst + x), t0);
583 _mm_storeu_si128( (__m128i*)(dst + x + 8), t1);
586 for( ; x <= width - 16; x += 16 )
588 __m128 x0, x1, y0, y1;
590 x0 = _mm_loadu_ps(S0 + x);
591 x1 = _mm_loadu_ps(S0 + x + 4);
592 y0 = _mm_loadu_ps(S1 + x);
593 y1 = _mm_loadu_ps(S1 + x + 4);
595 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
596 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
597 t0 = _mm_add_epi32(_mm_cvtps_epi32(x0), preshift);
598 t2 = _mm_add_epi32(_mm_cvtps_epi32(x1), preshift);
599 t0 = _mm_add_epi16(_mm_packs_epi32(t0, t2), postshift);
601 x0 = _mm_loadu_ps(S0 + x + 8);
602 x1 = _mm_loadu_ps(S0 + x + 12);
603 y0 = _mm_loadu_ps(S1 + x + 8);
604 y1 = _mm_loadu_ps(S1 + x + 12);
606 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
607 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
608 t1 = _mm_add_epi32(_mm_cvtps_epi32(x0), preshift);
609 t2 = _mm_add_epi32(_mm_cvtps_epi32(x1), preshift);
610 t1 = _mm_add_epi16(_mm_packs_epi32(t1, t2), postshift);
612 _mm_storeu_si128( (__m128i*)(dst + x), t0);
613 _mm_storeu_si128( (__m128i*)(dst + x + 8), t1);
616 for( ; x < width - 4; x += 4 )
620 x0 = _mm_loadu_ps(S0 + x);
621 y0 = _mm_loadu_ps(S1 + x);
623 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
624 t0 = _mm_add_epi32(_mm_cvtps_epi32(x0), preshift);
625 t0 = _mm_add_epi16(_mm_packs_epi32(t0, t0), postshift);
626 _mm_storel_epi64( (__m128i*)(dst + x), t0);
633 typedef VResizeLinearVec_32f16<SHRT_MIN> VResizeLinearVec_32f16u;
634 typedef VResizeLinearVec_32f16<0> VResizeLinearVec_32f16s;
636 struct VResizeLinearVec_32f
638 int operator()(const uchar** _src, uchar* _dst, const uchar* _beta, int width ) const
640 if( !checkHardwareSupport(CV_CPU_SSE) )
643 const float** src = (const float**)_src;
644 const float* beta = (const float*)_beta;
645 const float *S0 = src[0], *S1 = src[1];
646 float* dst = (float*)_dst;
649 __m128 b0 = _mm_set1_ps(beta[0]), b1 = _mm_set1_ps(beta[1]);
651 if( (((size_t)S0|(size_t)S1)&15) == 0 )
652 for( ; x <= width - 8; x += 8 )
654 __m128 x0, x1, y0, y1;
655 x0 = _mm_load_ps(S0 + x);
656 x1 = _mm_load_ps(S0 + x + 4);
657 y0 = _mm_load_ps(S1 + x);
658 y1 = _mm_load_ps(S1 + x + 4);
660 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
661 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
663 _mm_storeu_ps( dst + x, x0);
664 _mm_storeu_ps( dst + x + 4, x1);
667 for( ; x <= width - 8; x += 8 )
669 __m128 x0, x1, y0, y1;
670 x0 = _mm_loadu_ps(S0 + x);
671 x1 = _mm_loadu_ps(S0 + x + 4);
672 y0 = _mm_loadu_ps(S1 + x);
673 y1 = _mm_loadu_ps(S1 + x + 4);
675 x0 = _mm_add_ps(_mm_mul_ps(x0, b0), _mm_mul_ps(y0, b1));
676 x1 = _mm_add_ps(_mm_mul_ps(x1, b0), _mm_mul_ps(y1, b1));
678 _mm_storeu_ps( dst + x, x0);
679 _mm_storeu_ps( dst + x + 4, x1);
687 struct VResizeCubicVec_32s8u
689 int operator()(const uchar** _src, uchar* dst, const uchar* _beta, int width ) const
691 if( !checkHardwareSupport(CV_CPU_SSE2) )
694 const int** src = (const int**)_src;
695 const short* beta = (const short*)_beta;
696 const int *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];
698 float scale = 1.f/(INTER_RESIZE_COEF_SCALE*INTER_RESIZE_COEF_SCALE);
699 __m128 b0 = _mm_set1_ps(beta[0]*scale), b1 = _mm_set1_ps(beta[1]*scale),
700 b2 = _mm_set1_ps(beta[2]*scale), b3 = _mm_set1_ps(beta[3]*scale);
702 if( (((size_t)S0|(size_t)S1|(size_t)S2|(size_t)S3)&15) == 0 )
703 for( ; x <= width - 8; x += 8 )
705 __m128i x0, x1, y0, y1;
706 __m128 s0, s1, f0, f1;
707 x0 = _mm_load_si128((const __m128i*)(S0 + x));
708 x1 = _mm_load_si128((const __m128i*)(S0 + x + 4));
709 y0 = _mm_load_si128((const __m128i*)(S1 + x));
710 y1 = _mm_load_si128((const __m128i*)(S1 + x + 4));
712 s0 = _mm_mul_ps(_mm_cvtepi32_ps(x0), b0);
713 s1 = _mm_mul_ps(_mm_cvtepi32_ps(x1), b0);
714 f0 = _mm_mul_ps(_mm_cvtepi32_ps(y0), b1);
715 f1 = _mm_mul_ps(_mm_cvtepi32_ps(y1), b1);
716 s0 = _mm_add_ps(s0, f0);
717 s1 = _mm_add_ps(s1, f1);
719 x0 = _mm_load_si128((const __m128i*)(S2 + x));
720 x1 = _mm_load_si128((const __m128i*)(S2 + x + 4));
721 y0 = _mm_load_si128((const __m128i*)(S3 + x));
722 y1 = _mm_load_si128((const __m128i*)(S3 + x + 4));
724 f0 = _mm_mul_ps(_mm_cvtepi32_ps(x0), b2);
725 f1 = _mm_mul_ps(_mm_cvtepi32_ps(x1), b2);
726 s0 = _mm_add_ps(s0, f0);
727 s1 = _mm_add_ps(s1, f1);
728 f0 = _mm_mul_ps(_mm_cvtepi32_ps(y0), b3);
729 f1 = _mm_mul_ps(_mm_cvtepi32_ps(y1), b3);
730 s0 = _mm_add_ps(s0, f0);
731 s1 = _mm_add_ps(s1, f1);
733 x0 = _mm_cvtps_epi32(s0);
734 x1 = _mm_cvtps_epi32(s1);
736 x0 = _mm_packs_epi32(x0, x1);
737 _mm_storel_epi64( (__m128i*)(dst + x), _mm_packus_epi16(x0, x0));
740 for( ; x <= width - 8; x += 8 )
742 __m128i x0, x1, y0, y1;
743 __m128 s0, s1, f0, f1;
744 x0 = _mm_loadu_si128((const __m128i*)(S0 + x));
745 x1 = _mm_loadu_si128((const __m128i*)(S0 + x + 4));
746 y0 = _mm_loadu_si128((const __m128i*)(S1 + x));
747 y1 = _mm_loadu_si128((const __m128i*)(S1 + x + 4));
749 s0 = _mm_mul_ps(_mm_cvtepi32_ps(x0), b0);
750 s1 = _mm_mul_ps(_mm_cvtepi32_ps(x1), b0);
751 f0 = _mm_mul_ps(_mm_cvtepi32_ps(y0), b1);
752 f1 = _mm_mul_ps(_mm_cvtepi32_ps(y1), b1);
753 s0 = _mm_add_ps(s0, f0);
754 s1 = _mm_add_ps(s1, f1);
756 x0 = _mm_loadu_si128((const __m128i*)(S2 + x));
757 x1 = _mm_loadu_si128((const __m128i*)(S2 + x + 4));
758 y0 = _mm_loadu_si128((const __m128i*)(S3 + x));
759 y1 = _mm_loadu_si128((const __m128i*)(S3 + x + 4));
761 f0 = _mm_mul_ps(_mm_cvtepi32_ps(x0), b2);
762 f1 = _mm_mul_ps(_mm_cvtepi32_ps(x1), b2);
763 s0 = _mm_add_ps(s0, f0);
764 s1 = _mm_add_ps(s1, f1);
765 f0 = _mm_mul_ps(_mm_cvtepi32_ps(y0), b3);
766 f1 = _mm_mul_ps(_mm_cvtepi32_ps(y1), b3);
767 s0 = _mm_add_ps(s0, f0);
768 s1 = _mm_add_ps(s1, f1);
770 x0 = _mm_cvtps_epi32(s0);
771 x1 = _mm_cvtps_epi32(s1);
773 x0 = _mm_packs_epi32(x0, x1);
774 _mm_storel_epi64( (__m128i*)(dst + x), _mm_packus_epi16(x0, x0));
782 template<int shiftval> struct VResizeCubicVec_32f16
784 int operator()(const uchar** _src, uchar* _dst, const uchar* _beta, int width ) const
786 if( !checkHardwareSupport(CV_CPU_SSE2) )
789 const float** src = (const float**)_src;
790 const float* beta = (const float*)_beta;
791 const float *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];
792 ushort* dst = (ushort*)_dst;
794 __m128 b0 = _mm_set1_ps(beta[0]), b1 = _mm_set1_ps(beta[1]),
795 b2 = _mm_set1_ps(beta[2]), b3 = _mm_set1_ps(beta[3]);
796 __m128i preshift = _mm_set1_epi32(shiftval);
797 __m128i postshift = _mm_set1_epi16((short)shiftval);
799 for( ; x <= width - 8; x += 8 )
801 __m128 x0, x1, y0, y1, s0, s1;
803 x0 = _mm_loadu_ps(S0 + x);
804 x1 = _mm_loadu_ps(S0 + x + 4);
805 y0 = _mm_loadu_ps(S1 + x);
806 y1 = _mm_loadu_ps(S1 + x + 4);
808 s0 = _mm_mul_ps(x0, b0);
809 s1 = _mm_mul_ps(x1, b0);
810 y0 = _mm_mul_ps(y0, b1);
811 y1 = _mm_mul_ps(y1, b1);
812 s0 = _mm_add_ps(s0, y0);
813 s1 = _mm_add_ps(s1, y1);
815 x0 = _mm_loadu_ps(S2 + x);
816 x1 = _mm_loadu_ps(S2 + x + 4);
817 y0 = _mm_loadu_ps(S3 + x);
818 y1 = _mm_loadu_ps(S3 + x + 4);
820 x0 = _mm_mul_ps(x0, b2);
821 x1 = _mm_mul_ps(x1, b2);
822 y0 = _mm_mul_ps(y0, b3);
823 y1 = _mm_mul_ps(y1, b3);
824 s0 = _mm_add_ps(s0, x0);
825 s1 = _mm_add_ps(s1, x1);
826 s0 = _mm_add_ps(s0, y0);
827 s1 = _mm_add_ps(s1, y1);
829 t0 = _mm_add_epi32(_mm_cvtps_epi32(s0), preshift);
830 t1 = _mm_add_epi32(_mm_cvtps_epi32(s1), preshift);
832 t0 = _mm_add_epi16(_mm_packs_epi32(t0, t1), postshift);
833 _mm_storeu_si128( (__m128i*)(dst + x), t0);
840 typedef VResizeCubicVec_32f16<SHRT_MIN> VResizeCubicVec_32f16u;
841 typedef VResizeCubicVec_32f16<0> VResizeCubicVec_32f16s;
843 struct VResizeCubicVec_32f
845 int operator()(const uchar** _src, uchar* _dst, const uchar* _beta, int width ) const
847 if( !checkHardwareSupport(CV_CPU_SSE) )
850 const float** src = (const float**)_src;
851 const float* beta = (const float*)_beta;
852 const float *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];
853 float* dst = (float*)_dst;
855 __m128 b0 = _mm_set1_ps(beta[0]), b1 = _mm_set1_ps(beta[1]),
856 b2 = _mm_set1_ps(beta[2]), b3 = _mm_set1_ps(beta[3]);
858 for( ; x <= width - 8; x += 8 )
860 __m128 x0, x1, y0, y1, s0, s1;
861 x0 = _mm_loadu_ps(S0 + x);
862 x1 = _mm_loadu_ps(S0 + x + 4);
863 y0 = _mm_loadu_ps(S1 + x);
864 y1 = _mm_loadu_ps(S1 + x + 4);
866 s0 = _mm_mul_ps(x0, b0);
867 s1 = _mm_mul_ps(x1, b0);
868 y0 = _mm_mul_ps(y0, b1);
869 y1 = _mm_mul_ps(y1, b1);
870 s0 = _mm_add_ps(s0, y0);
871 s1 = _mm_add_ps(s1, y1);
873 x0 = _mm_loadu_ps(S2 + x);
874 x1 = _mm_loadu_ps(S2 + x + 4);
875 y0 = _mm_loadu_ps(S3 + x);
876 y1 = _mm_loadu_ps(S3 + x + 4);
878 x0 = _mm_mul_ps(x0, b2);
879 x1 = _mm_mul_ps(x1, b2);
880 y0 = _mm_mul_ps(y0, b3);
881 y1 = _mm_mul_ps(y1, b3);
882 s0 = _mm_add_ps(s0, x0);
883 s1 = _mm_add_ps(s1, x1);
884 s0 = _mm_add_ps(s0, y0);
885 s1 = _mm_add_ps(s1, y1);
887 _mm_storeu_ps( dst + x, s0);
888 _mm_storeu_ps( dst + x + 4, s1);
897 typedef VResizeNoVec VResizeLinearVec_32s8u;
898 typedef VResizeNoVec VResizeLinearVec_32f16u;
899 typedef VResizeNoVec VResizeLinearVec_32f16s;
900 typedef VResizeNoVec VResizeLinearVec_32f;
902 typedef VResizeNoVec VResizeCubicVec_32s8u;
903 typedef VResizeNoVec VResizeCubicVec_32f16u;
904 typedef VResizeNoVec VResizeCubicVec_32f16s;
905 typedef VResizeNoVec VResizeCubicVec_32f;
909 typedef HResizeNoVec HResizeLinearVec_8u32s;
910 typedef HResizeNoVec HResizeLinearVec_16u32f;
911 typedef HResizeNoVec HResizeLinearVec_16s32f;
912 typedef HResizeNoVec HResizeLinearVec_32f;
913 typedef HResizeNoVec HResizeLinearVec_64f;
916 template<typename T, typename WT, typename AT, int ONE, class VecOp>
919 typedef T value_type;
921 typedef AT alpha_type;
923 void operator()(const T** src, WT** dst, int count,
924 const int* xofs, const AT* alpha,
925 int swidth, int dwidth, int cn, int xmin, int xmax ) const
930 int dx0 = vecOp((const uchar**)src, (uchar**)dst, count,
931 xofs, (const uchar*)alpha, swidth, dwidth, cn, xmin, xmax );
933 for( k = 0; k <= count - 2; k++ )
935 const T *S0 = src[k], *S1 = src[k+1];
936 WT *D0 = dst[k], *D1 = dst[k+1];
937 for( dx = dx0; dx < xmax; dx++ )
940 WT a0 = alpha[dx*2], a1 = alpha[dx*2+1];
941 WT t0 = S0[sx]*a0 + S0[sx + cn]*a1;
942 WT t1 = S1[sx]*a0 + S1[sx + cn]*a1;
943 D0[dx] = t0; D1[dx] = t1;
946 for( ; dx < dwidth; dx++ )
949 D0[dx] = WT(S0[sx]*ONE); D1[dx] = WT(S1[sx]*ONE);
953 for( ; k < count; k++ )
957 for( dx = 0; dx < xmax; dx++ )
960 D[dx] = S[sx]*alpha[dx*2] + S[sx+cn]*alpha[dx*2+1];
963 for( ; dx < dwidth; dx++ )
964 D[dx] = WT(S[xofs[dx]]*ONE);
970 template<typename T, typename WT, typename AT, class CastOp, class VecOp>
973 typedef T value_type;
975 typedef AT alpha_type;
977 void operator()(const WT** src, T* dst, const AT* beta, int width ) const
979 WT b0 = beta[0], b1 = beta[1];
980 const WT *S0 = src[0], *S1 = src[1];
984 int x = vecOp((const uchar**)src, (uchar*)dst, (const uchar*)beta, width);
985 #if CV_ENABLE_UNROLLED
986 for( ; x <= width - 4; x += 4 )
989 t0 = S0[x]*b0 + S1[x]*b1;
990 t1 = S0[x+1]*b0 + S1[x+1]*b1;
991 dst[x] = castOp(t0); dst[x+1] = castOp(t1);
992 t0 = S0[x+2]*b0 + S1[x+2]*b1;
993 t1 = S0[x+3]*b0 + S1[x+3]*b1;
994 dst[x+2] = castOp(t0); dst[x+3] = castOp(t1);
997 for( ; x < width; x++ )
998 dst[x] = castOp(S0[x]*b0 + S1[x]*b1);
1003 struct VResizeLinear<uchar, int, short, FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS*2>, VResizeLinearVec_32s8u>
1005 typedef uchar value_type;
1006 typedef int buf_type;
1007 typedef short alpha_type;
1009 void operator()(const buf_type** src, value_type* dst, const alpha_type* beta, int width ) const
1011 alpha_type b0 = beta[0], b1 = beta[1];
1012 const buf_type *S0 = src[0], *S1 = src[1];
1013 VResizeLinearVec_32s8u vecOp;
1015 int x = vecOp((const uchar**)src, (uchar*)dst, (const uchar*)beta, width);
1016 #if CV_ENABLE_UNROLLED
1017 for( ; x <= width - 4; x += 4 )
1019 dst[x+0] = uchar(( ((b0 * (S0[x+0] >> 4)) >> 16) + ((b1 * (S1[x+0] >> 4)) >> 16) + 2)>>2);
1020 dst[x+1] = uchar(( ((b0 * (S0[x+1] >> 4)) >> 16) + ((b1 * (S1[x+1] >> 4)) >> 16) + 2)>>2);
1021 dst[x+2] = uchar(( ((b0 * (S0[x+2] >> 4)) >> 16) + ((b1 * (S1[x+2] >> 4)) >> 16) + 2)>>2);
1022 dst[x+3] = uchar(( ((b0 * (S0[x+3] >> 4)) >> 16) + ((b1 * (S1[x+3] >> 4)) >> 16) + 2)>>2);
1025 for( ; x < width; x++ )
1026 dst[x] = uchar(( ((b0 * (S0[x] >> 4)) >> 16) + ((b1 * (S1[x] >> 4)) >> 16) + 2)>>2);
1031 template<typename T, typename WT, typename AT>
1034 typedef T value_type;
1035 typedef WT buf_type;
1036 typedef AT alpha_type;
1038 void operator()(const T** src, WT** dst, int count,
1039 const int* xofs, const AT* alpha,
1040 int swidth, int dwidth, int cn, int xmin, int xmax ) const
1042 for( int k = 0; k < count; k++ )
1044 const T *S = src[k];
1046 int dx = 0, limit = xmin;
1049 for( ; dx < limit; dx++, alpha += 4 )
1051 int j, sx = xofs[dx] - cn;
1053 for( j = 0; j < 4; j++ )
1055 int sxj = sx + j*cn;
1056 if( (unsigned)sxj >= (unsigned)swidth )
1060 while( sxj >= swidth )
1063 v += S[sxj]*alpha[j];
1067 if( limit == dwidth )
1069 for( ; dx < xmax; dx++, alpha += 4 )
1072 D[dx] = S[sx-cn]*alpha[0] + S[sx]*alpha[1] +
1073 S[sx+cn]*alpha[2] + S[sx+cn*2]*alpha[3];
1083 template<typename T, typename WT, typename AT, class CastOp, class VecOp>
1086 typedef T value_type;
1087 typedef WT buf_type;
1088 typedef AT alpha_type;
1090 void operator()(const WT** src, T* dst, const AT* beta, int width ) const
1092 WT b0 = beta[0], b1 = beta[1], b2 = beta[2], b3 = beta[3];
1093 const WT *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];
1097 int x = vecOp((const uchar**)src, (uchar*)dst, (const uchar*)beta, width);
1098 for( ; x < width; x++ )
1099 dst[x] = castOp(S0[x]*b0 + S1[x]*b1 + S2[x]*b2 + S3[x]*b3);
1104 template<typename T, typename WT, typename AT>
1105 struct HResizeLanczos4
1107 typedef T value_type;
1108 typedef WT buf_type;
1109 typedef AT alpha_type;
1111 void operator()(const T** src, WT** dst, int count,
1112 const int* xofs, const AT* alpha,
1113 int swidth, int dwidth, int cn, int xmin, int xmax ) const
1115 for( int k = 0; k < count; k++ )
1117 const T *S = src[k];
1119 int dx = 0, limit = xmin;
1122 for( ; dx < limit; dx++, alpha += 8 )
1124 int j, sx = xofs[dx] - cn*3;
1126 for( j = 0; j < 8; j++ )
1128 int sxj = sx + j*cn;
1129 if( (unsigned)sxj >= (unsigned)swidth )
1133 while( sxj >= swidth )
1136 v += S[sxj]*alpha[j];
1140 if( limit == dwidth )
1142 for( ; dx < xmax; dx++, alpha += 8 )
1145 D[dx] = S[sx-cn*3]*alpha[0] + S[sx-cn*2]*alpha[1] +
1146 S[sx-cn]*alpha[2] + S[sx]*alpha[3] +
1147 S[sx+cn]*alpha[4] + S[sx+cn*2]*alpha[5] +
1148 S[sx+cn*3]*alpha[6] + S[sx+cn*4]*alpha[7];
1158 template<typename T, typename WT, typename AT, class CastOp, class VecOp>
1159 struct VResizeLanczos4
1161 typedef T value_type;
1162 typedef WT buf_type;
1163 typedef AT alpha_type;
1165 void operator()(const WT** src, T* dst, const AT* beta, int width ) const
1169 int k, x = vecOp((const uchar**)src, (uchar*)dst, (const uchar*)beta, width);
1170 #if CV_ENABLE_UNROLLED
1171 for( ; x <= width - 4; x += 4 )
1174 const WT* S = src[0];
1175 WT s0 = S[x]*b, s1 = S[x+1]*b, s2 = S[x+2]*b, s3 = S[x+3]*b;
1177 for( k = 1; k < 8; k++ )
1179 b = beta[k]; S = src[k];
1180 s0 += S[x]*b; s1 += S[x+1]*b;
1181 s2 += S[x+2]*b; s3 += S[x+3]*b;
1184 dst[x] = castOp(s0); dst[x+1] = castOp(s1);
1185 dst[x+2] = castOp(s2); dst[x+3] = castOp(s3);
1188 for( ; x < width; x++ )
1190 dst[x] = castOp(src[0][x]*beta[0] + src[1][x]*beta[1] +
1191 src[2][x]*beta[2] + src[3][x]*beta[3] + src[4][x]*beta[4] +
1192 src[5][x]*beta[5] + src[6][x]*beta[6] + src[7][x]*beta[7]);
1198 static inline int clip(int x, int a, int b)
1200 return x >= a ? (x < b ? x : b-1) : a;
1203 static const int MAX_ESIZE=16;
1205 template <typename HResize, typename VResize>
1206 class resizeGeneric_Invoker :
1207 public ParallelLoopBody
1210 typedef typename HResize::value_type T;
1211 typedef typename HResize::buf_type WT;
1212 typedef typename HResize::alpha_type AT;
1214 resizeGeneric_Invoker(const Mat& _src, Mat &_dst, const int *_xofs, const int *_yofs,
1215 const AT* _alpha, const AT* __beta, const Size& _ssize, const Size &_dsize,
1216 int _ksize, int _xmin, int _xmax) :
1217 ParallelLoopBody(), src(_src), dst(_dst), xofs(_xofs), yofs(_yofs),
1218 alpha(_alpha), _beta(__beta), ssize(_ssize), dsize(_dsize),
1219 ksize(_ksize), xmin(_xmin), xmax(_xmax)
1223 virtual void operator() (const Range& range) const
1225 int dy, cn = src.channels();
1229 int bufstep = (int)alignSize(dsize.width, 16);
1230 AutoBuffer<WT> _buffer(bufstep*ksize);
1231 const T* srows[MAX_ESIZE]={0};
1232 WT* rows[MAX_ESIZE]={0};
1233 int prev_sy[MAX_ESIZE];
1235 for(int k = 0; k < ksize; k++ )
1238 rows[k] = (WT*)_buffer + bufstep*k;
1241 const AT* beta = _beta + ksize * range.start;
1243 for( dy = range.start; dy < range.end; dy++, beta += ksize )
1245 int sy0 = yofs[dy], k0=ksize, k1=0, ksize2 = ksize/2;
1247 for(int k = 0; k < ksize; k++ )
1249 int sy = clip(sy0 - ksize2 + 1 + k, 0, ssize.height);
1250 for( k1 = std::max(k1, k); k1 < ksize; k1++ )
1252 if( sy == prev_sy[k1] ) // if the sy-th row has been computed already, reuse it.
1255 memcpy( rows[k], rows[k1], bufstep*sizeof(rows[0][0]) );
1260 k0 = std::min(k0, k); // remember the first row that needs to be computed
1261 srows[k] = (T*)(src.data + src.step*sy);
1266 hresize( (const T**)(srows + k0), (WT**)(rows + k0), ksize - k0, xofs, (const AT*)(alpha),
1267 ssize.width, dsize.width, cn, xmin, xmax );
1268 vresize( (const WT**)rows, (T*)(dst.data + dst.step*dy), beta, dsize.width );
1275 const int* xofs, *yofs;
1276 const AT* alpha, *_beta;
1278 int ksize, xmin, xmax;
1281 template<class HResize, class VResize>
1282 static void resizeGeneric_( const Mat& src, Mat& dst,
1283 const int* xofs, const void* _alpha,
1284 const int* yofs, const void* _beta,
1285 int xmin, int xmax, int ksize )
1287 typedef typename HResize::alpha_type AT;
1289 const AT* beta = (const AT*)_beta;
1290 Size ssize = src.size(), dsize = dst.size();
1291 int cn = src.channels();
1296 // image resize is a separable operation. In case of not too strong
1298 Range range(0, dsize.height);
1299 resizeGeneric_Invoker<HResize, VResize> invoker(src, dst, xofs, yofs, (const AT*)_alpha, beta,
1300 ssize, dsize, ksize, xmin, xmax);
1301 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
1304 template <typename T, typename WT>
1305 struct ResizeAreaFastNoVec
1307 ResizeAreaFastNoVec(int, int) { }
1308 ResizeAreaFastNoVec(int, int, int, int) { }
1309 int operator() (const T*, T*, int) const
1314 class ResizeAreaFastVec_SIMD_8u
1317 ResizeAreaFastVec_SIMD_8u(int _cn, int _step) :
1318 cn(_cn), step(_step)
1320 use_simd = checkHardwareSupport(CV_CPU_SSE2);
1323 int operator() (const uchar* S, uchar* D, int w) const
1329 const uchar* S0 = S;
1330 const uchar* S1 = S0 + step;
1331 __m128i zero = _mm_setzero_si128();
1332 __m128i delta2 = _mm_set1_epi16(2);
1336 __m128i masklow = _mm_set1_epi16(0x00ff);
1337 for ( ; dx <= w - 8; dx += 8, S0 += 16, S1 += 16, D += 8)
1339 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1340 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1342 __m128i s0 = _mm_add_epi16(_mm_srli_epi16(r0, 8), _mm_and_si128(r0, masklow));
1343 __m128i s1 = _mm_add_epi16(_mm_srli_epi16(r1, 8), _mm_and_si128(r1, masklow));
1344 s0 = _mm_add_epi16(_mm_add_epi16(s0, s1), delta2);
1345 s0 = _mm_packus_epi16(_mm_srli_epi16(s0, 2), zero);
1347 _mm_storel_epi64((__m128i*)D, s0);
1351 for ( ; dx <= w - 11; dx += 6, S0 += 12, S1 += 12, D += 6)
1353 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1354 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1356 __m128i r0_16l = _mm_unpacklo_epi8(r0, zero);
1357 __m128i r0_16h = _mm_unpacklo_epi8(_mm_srli_si128(r0, 6), zero);
1358 __m128i r1_16l = _mm_unpacklo_epi8(r1, zero);
1359 __m128i r1_16h = _mm_unpacklo_epi8(_mm_srli_si128(r1, 6), zero);
1361 __m128i s0 = _mm_add_epi16(r0_16l, _mm_srli_si128(r0_16l, 6));
1362 __m128i s1 = _mm_add_epi16(r1_16l, _mm_srli_si128(r1_16l, 6));
1363 s0 = _mm_add_epi16(s1, _mm_add_epi16(s0, delta2));
1364 s0 = _mm_packus_epi16(_mm_srli_epi16(s0, 2), zero);
1365 _mm_storel_epi64((__m128i*)D, s0);
1367 s0 = _mm_add_epi16(r0_16h, _mm_srli_si128(r0_16h, 6));
1368 s1 = _mm_add_epi16(r1_16h, _mm_srli_si128(r1_16h, 6));
1369 s0 = _mm_add_epi16(s1, _mm_add_epi16(s0, delta2));
1370 s0 = _mm_packus_epi16(_mm_srli_epi16(s0, 2), zero);
1371 _mm_storel_epi64((__m128i*)(D+3), s0);
1376 int v[] = { 0, 0, -1, -1 };
1377 __m128i mask = _mm_loadu_si128((const __m128i*)v);
1379 for ( ; dx <= w - 8; dx += 8, S0 += 16, S1 += 16, D += 8)
1381 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1382 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1384 __m128i r0_16l = _mm_unpacklo_epi8(r0, zero);
1385 __m128i r0_16h = _mm_unpackhi_epi8(r0, zero);
1386 __m128i r1_16l = _mm_unpacklo_epi8(r1, zero);
1387 __m128i r1_16h = _mm_unpackhi_epi8(r1, zero);
1389 __m128i s0 = _mm_add_epi16(r0_16l, _mm_srli_si128(r0_16l, 8));
1390 __m128i s1 = _mm_add_epi16(r1_16l, _mm_srli_si128(r1_16l, 8));
1391 s0 = _mm_add_epi16(s1, _mm_add_epi16(s0, delta2));
1392 __m128i res0 = _mm_srli_epi16(s0, 2);
1394 s0 = _mm_add_epi16(r0_16h, _mm_srli_si128(r0_16h, 8));
1395 s1 = _mm_add_epi16(r1_16h, _mm_srli_si128(r1_16h, 8));
1396 s0 = _mm_add_epi16(s1, _mm_add_epi16(s0, delta2));
1397 __m128i res1 = _mm_srli_epi16(s0, 2);
1398 s0 = _mm_packus_epi16(_mm_or_si128(_mm_andnot_si128(mask, res0),
1399 _mm_and_si128(mask, _mm_slli_si128(res1, 8))), zero);
1400 _mm_storel_epi64((__m128i*)(D), s0);
1413 class ResizeAreaFastVec_SIMD_16u
1416 ResizeAreaFastVec_SIMD_16u(int _cn, int _step) :
1417 cn(_cn), step(_step)
1419 use_simd = checkHardwareSupport(CV_CPU_SSE2);
1422 int operator() (const ushort* S, ushort* D, int w) const
1428 const ushort* S0 = (const ushort*)S;
1429 const ushort* S1 = (const ushort*)((const uchar*)(S) + step);
1430 __m128i masklow = _mm_set1_epi32(0x0000ffff);
1431 __m128i zero = _mm_setzero_si128();
1432 __m128i delta2 = _mm_set1_epi32(2);
1434 #define _mm_packus_epi32(a, zero) _mm_packs_epi32(_mm_srai_epi32(_mm_slli_epi32(a, 16), 16), zero)
1438 for ( ; dx <= w - 4; dx += 4, S0 += 8, S1 += 8, D += 4)
1440 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1441 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1443 __m128i s0 = _mm_add_epi32(_mm_srli_epi32(r0, 16), _mm_and_si128(r0, masklow));
1444 __m128i s1 = _mm_add_epi32(_mm_srli_epi32(r1, 16), _mm_and_si128(r1, masklow));
1445 s0 = _mm_add_epi32(_mm_add_epi32(s0, s1), delta2);
1446 s0 = _mm_srli_epi32(s0, 2);
1447 s0 = _mm_packus_epi32(s0, zero);
1449 _mm_storel_epi64((__m128i*)D, s0);
1453 for ( ; dx <= w - 4; dx += 3, S0 += 6, S1 += 6, D += 3)
1455 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1456 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1458 __m128i r0_16l = _mm_unpacklo_epi16(r0, zero);
1459 __m128i r0_16h = _mm_unpacklo_epi16(_mm_srli_si128(r0, 6), zero);
1460 __m128i r1_16l = _mm_unpacklo_epi16(r1, zero);
1461 __m128i r1_16h = _mm_unpacklo_epi16(_mm_srli_si128(r1, 6), zero);
1463 __m128i s0 = _mm_add_epi32(r0_16l, r0_16h);
1464 __m128i s1 = _mm_add_epi32(r1_16l, r1_16h);
1465 s0 = _mm_add_epi32(delta2, _mm_add_epi32(s0, s1));
1466 s0 = _mm_packus_epi32(_mm_srli_epi32(s0, 2), zero);
1467 _mm_storel_epi64((__m128i*)D, s0);
1472 for ( ; dx <= w - 4; dx += 4, S0 += 8, S1 += 8, D += 4)
1474 __m128i r0 = _mm_loadu_si128((const __m128i*)S0);
1475 __m128i r1 = _mm_loadu_si128((const __m128i*)S1);
1477 __m128i r0_32l = _mm_unpacklo_epi16(r0, zero);
1478 __m128i r0_32h = _mm_unpackhi_epi16(r0, zero);
1479 __m128i r1_32l = _mm_unpacklo_epi16(r1, zero);
1480 __m128i r1_32h = _mm_unpackhi_epi16(r1, zero);
1482 __m128i s0 = _mm_add_epi32(r0_32l, r0_32h);
1483 __m128i s1 = _mm_add_epi32(r1_32l, r1_32h);
1484 s0 = _mm_add_epi32(s1, _mm_add_epi32(s0, delta2));
1485 s0 = _mm_packus_epi32(_mm_srli_epi32(s0, 2), zero);
1486 _mm_storel_epi64((__m128i*)D, s0);
1490 #undef _mm_packus_epi32
1502 typedef ResizeAreaFastNoVec<uchar, uchar> ResizeAreaFastVec_SIMD_8u;
1503 typedef ResizeAreaFastNoVec<ushort, ushort> ResizeAreaFastVec_SIMD_16u;
1506 template<typename T, typename SIMDVecOp>
1507 struct ResizeAreaFastVec
1509 ResizeAreaFastVec(int _scale_x, int _scale_y, int _cn, int _step) :
1510 scale_x(_scale_x), scale_y(_scale_y), cn(_cn), step(_step), vecOp(_cn, _step)
1512 fast_mode = scale_x == 2 && scale_y == 2 && (cn == 1 || cn == 3 || cn == 4);
1515 int operator() (const T* S, T* D, int w) const
1520 const T* nextS = (const T*)((const uchar*)S + step);
1521 int dx = vecOp(S, D, w);
1524 for( ; dx < w; ++dx )
1527 D[dx] = (T)((S[index] + S[index+1] + nextS[index] + nextS[index+1] + 2) >> 2);
1530 for( ; dx < w; dx += 3 )
1533 D[dx] = (T)((S[index] + S[index+3] + nextS[index] + nextS[index+3] + 2) >> 2);
1534 D[dx+1] = (T)((S[index+1] + S[index+4] + nextS[index+1] + nextS[index+4] + 2) >> 2);
1535 D[dx+2] = (T)((S[index+2] + S[index+5] + nextS[index+2] + nextS[index+5] + 2) >> 2);
1540 for( ; dx < w; dx += 4 )
1543 D[dx] = (T)((S[index] + S[index+4] + nextS[index] + nextS[index+4] + 2) >> 2);
1544 D[dx+1] = (T)((S[index+1] + S[index+5] + nextS[index+1] + nextS[index+5] + 2) >> 2);
1545 D[dx+2] = (T)((S[index+2] + S[index+6] + nextS[index+2] + nextS[index+6] + 2) >> 2);
1546 D[dx+3] = (T)((S[index+3] + S[index+7] + nextS[index+3] + nextS[index+7] + 2) >> 2);
1554 int scale_x, scale_y;
1561 template <typename T, typename WT, typename VecOp>
1562 class resizeAreaFast_Invoker :
1563 public ParallelLoopBody
1566 resizeAreaFast_Invoker(const Mat &_src, Mat &_dst,
1567 int _scale_x, int _scale_y, const int* _ofs, const int* _xofs) :
1568 ParallelLoopBody(), src(_src), dst(_dst), scale_x(_scale_x),
1569 scale_y(_scale_y), ofs(_ofs), xofs(_xofs)
1573 virtual void operator() (const Range& range) const
1575 Size ssize = src.size(), dsize = dst.size();
1576 int cn = src.channels();
1577 int area = scale_x*scale_y;
1578 float scale = 1.f/(area);
1579 int dwidth1 = (ssize.width/scale_x)*cn;
1584 VecOp vop(scale_x, scale_y, src.channels(), (int)src.step/*, area_ofs*/);
1586 for( dy = range.start; dy < range.end; dy++ )
1588 T* D = (T*)(dst.data + dst.step*dy);
1589 int sy0 = dy*scale_y;
1590 int w = sy0 + scale_y <= ssize.height ? dwidth1 : 0;
1592 if( sy0 >= ssize.height )
1594 for( dx = 0; dx < dsize.width; dx++ )
1599 dx = vop((const T*)(src.data + src.step * sy0), D, w);
1600 for( ; dx < w; dx++ )
1602 const T* S = (const T*)(src.data + src.step * sy0) + xofs[dx];
1605 #if CV_ENABLE_UNROLLED
1606 for( ; k <= area - 4; k += 4 )
1607 sum += S[ofs[k]] + S[ofs[k+1]] + S[ofs[k+2]] + S[ofs[k+3]];
1609 for( ; k < area; k++ )
1612 D[dx] = saturate_cast<T>(sum * scale);
1615 for( ; dx < dsize.width; dx++ )
1618 int count = 0, sx0 = xofs[dx];
1619 if( sx0 >= ssize.width )
1622 for( int sy = 0; sy < scale_y; sy++ )
1624 if( sy0 + sy >= ssize.height )
1626 const T* S = (const T*)(src.data + src.step*(sy0 + sy)) + sx0;
1627 for( int sx = 0; sx < scale_x*cn; sx += cn )
1629 if( sx0 + sx >= ssize.width )
1636 D[dx] = saturate_cast<T>((float)sum/count);
1644 int scale_x, scale_y;
1645 const int *ofs, *xofs;
1648 template<typename T, typename WT, typename VecOp>
1649 static void resizeAreaFast_( const Mat& src, Mat& dst, const int* ofs, const int* xofs,
1650 int scale_x, int scale_y )
1652 Range range(0, dst.rows);
1653 resizeAreaFast_Invoker<T, WT, VecOp> invoker(src, dst, scale_x,
1654 scale_y, ofs, xofs);
1655 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
1658 struct DecimateAlpha
1665 template<typename T, typename WT> class ResizeArea_Invoker :
1666 public ParallelLoopBody
1669 ResizeArea_Invoker( const Mat& _src, Mat& _dst,
1670 const DecimateAlpha* _xtab, int _xtab_size,
1671 const DecimateAlpha* _ytab, int _ytab_size,
1672 const int* _tabofs )
1677 xtab_size0 = _xtab_size;
1679 ytab_size = _ytab_size;
1683 virtual void operator() (const Range& range) const
1685 Size dsize = dst->size();
1686 int cn = dst->channels();
1688 AutoBuffer<WT> _buffer(dsize.width*2);
1689 const DecimateAlpha* xtab = xtab0;
1690 int xtab_size = xtab_size0;
1691 WT *buf = _buffer, *sum = buf + dsize.width;
1692 int j_start = tabofs[range.start], j_end = tabofs[range.end], j, k, dx, prev_dy = ytab[j_start].di;
1694 for( dx = 0; dx < dsize.width; dx++ )
1697 for( j = j_start; j < j_end; j++ )
1699 WT beta = ytab[j].alpha;
1700 int dy = ytab[j].di;
1701 int sy = ytab[j].si;
1704 const T* S = (const T*)(src->data + src->step*sy);
1705 for( dx = 0; dx < dsize.width; dx++ )
1709 for( k = 0; k < xtab_size; k++ )
1711 int dxn = xtab[k].di;
1712 WT alpha = xtab[k].alpha;
1713 buf[dxn] += S[xtab[k].si]*alpha;
1716 for( k = 0; k < xtab_size; k++ )
1718 int sxn = xtab[k].si;
1719 int dxn = xtab[k].di;
1720 WT alpha = xtab[k].alpha;
1721 WT t0 = buf[dxn] + S[sxn]*alpha;
1722 WT t1 = buf[dxn+1] + S[sxn+1]*alpha;
1723 buf[dxn] = t0; buf[dxn+1] = t1;
1726 for( k = 0; k < xtab_size; k++ )
1728 int sxn = xtab[k].si;
1729 int dxn = xtab[k].di;
1730 WT alpha = xtab[k].alpha;
1731 WT t0 = buf[dxn] + S[sxn]*alpha;
1732 WT t1 = buf[dxn+1] + S[sxn+1]*alpha;
1733 WT t2 = buf[dxn+2] + S[sxn+2]*alpha;
1734 buf[dxn] = t0; buf[dxn+1] = t1; buf[dxn+2] = t2;
1738 for( k = 0; k < xtab_size; k++ )
1740 int sxn = xtab[k].si;
1741 int dxn = xtab[k].di;
1742 WT alpha = xtab[k].alpha;
1743 WT t0 = buf[dxn] + S[sxn]*alpha;
1744 WT t1 = buf[dxn+1] + S[sxn+1]*alpha;
1745 buf[dxn] = t0; buf[dxn+1] = t1;
1746 t0 = buf[dxn+2] + S[sxn+2]*alpha;
1747 t1 = buf[dxn+3] + S[sxn+3]*alpha;
1748 buf[dxn+2] = t0; buf[dxn+3] = t1;
1753 for( k = 0; k < xtab_size; k++ )
1755 int sxn = xtab[k].si;
1756 int dxn = xtab[k].di;
1757 WT alpha = xtab[k].alpha;
1758 for( int c = 0; c < cn; c++ )
1759 buf[dxn + c] += S[sxn + c]*alpha;
1766 T* D = (T*)(dst->data + dst->step*prev_dy);
1768 for( dx = 0; dx < dsize.width; dx++ )
1770 D[dx] = saturate_cast<T>(sum[dx]);
1771 sum[dx] = beta*buf[dx];
1777 for( dx = 0; dx < dsize.width; dx++ )
1778 sum[dx] += beta*buf[dx];
1783 T* D = (T*)(dst->data + dst->step*prev_dy);
1784 for( dx = 0; dx < dsize.width; dx++ )
1785 D[dx] = saturate_cast<T>(sum[dx]);
1792 const DecimateAlpha* xtab0;
1793 const DecimateAlpha* ytab;
1794 int xtab_size0, ytab_size;
1799 template <typename T, typename WT>
1800 static void resizeArea_( const Mat& src, Mat& dst,
1801 const DecimateAlpha* xtab, int xtab_size,
1802 const DecimateAlpha* ytab, int ytab_size,
1805 parallel_for_(Range(0, dst.rows),
1806 ResizeArea_Invoker<T, WT>(src, dst, xtab, xtab_size, ytab, ytab_size, tabofs),
1807 dst.total()/((double)(1 << 16)));
1811 typedef void (*ResizeFunc)( const Mat& src, Mat& dst,
1812 const int* xofs, const void* alpha,
1813 const int* yofs, const void* beta,
1814 int xmin, int xmax, int ksize );
1816 typedef void (*ResizeAreaFastFunc)( const Mat& src, Mat& dst,
1817 const int* ofs, const int *xofs,
1818 int scale_x, int scale_y );
1820 typedef void (*ResizeAreaFunc)( const Mat& src, Mat& dst,
1821 const DecimateAlpha* xtab, int xtab_size,
1822 const DecimateAlpha* ytab, int ytab_size,
1826 static int computeResizeAreaTab( int ssize, int dsize, int cn, double scale, DecimateAlpha* tab )
1829 for(int dx = 0; dx < dsize; dx++ )
1831 double fsx1 = dx * scale;
1832 double fsx2 = fsx1 + scale;
1833 double cellWidth = std::min(scale, ssize - fsx1);
1835 int sx1 = cvCeil(fsx1), sx2 = cvFloor(fsx2);
1837 sx2 = std::min(sx2, ssize - 1);
1838 sx1 = std::min(sx1, sx2);
1840 if( sx1 - fsx1 > 1e-3 )
1842 assert( k < ssize*2 );
1843 tab[k].di = dx * cn;
1844 tab[k].si = (sx1 - 1) * cn;
1845 tab[k++].alpha = (float)((sx1 - fsx1) / cellWidth);
1848 for(int sx = sx1; sx < sx2; sx++ )
1850 assert( k < ssize*2 );
1851 tab[k].di = dx * cn;
1852 tab[k].si = sx * cn;
1853 tab[k++].alpha = float(1.0 / cellWidth);
1856 if( fsx2 - sx2 > 1e-3 )
1858 assert( k < ssize*2 );
1859 tab[k].di = dx * cn;
1860 tab[k].si = sx2 * cn;
1861 tab[k++].alpha = (float)(std::min(std::min(fsx2 - sx2, 1.), cellWidth) / cellWidth);
1867 #define CHECK_FUNC(FUNC) if( FUNC==0 ) { *ok = false; return;}
1868 #define CHECK_STATUS(STATUS) if( STATUS!=ippStsNoErr ) { *ok = false; return;}
1870 #define SET_IPP_RESIZE_LINEAR_FUNC_PTR(TYPE, CN) \
1871 func = (ippiResizeFunc)ippiResizeLinear_##TYPE##_##CN##R; CHECK_FUNC(func);\
1872 status = ippiResizeGetSize_##TYPE(srcSize, dstSize, (IppiInterpolationType)mode, 0, &specSize, &initSize); CHECK_STATUS(status)\
1873 specBuf.allocate(specSize);\
1874 pSpec = (uchar*)specBuf;\
1875 status = ippiResizeLinearInit_##TYPE(srcSize, dstSize, (IppiResizeSpec_32f*)pSpec); CHECK_STATUS(status);
1877 #define SET_IPP_RESIZE_LINEAR_FUNC_64_PTR(TYPE, CN) \
1878 func = (ippiResizeFunc)ippiResizeLinear_##TYPE##_##CN##R; CHECK_FUNC(func);\
1879 status = ippiResizeGetSize_##TYPE(srcSize, dstSize, (IppiInterpolationType)mode, 0, &specSize, &initSize); CHECK_STATUS(status)\
1880 specBuf.allocate(specSize);\
1881 pSpec = (uchar*)specBuf;\
1882 status = ippiResizeLinearInit_##TYPE(srcSize, dstSize, (IppiResizeSpec_64f*)pSpec); CHECK_STATUS(status);
1884 #define SET_IPP_RESIZE_CUBIC_FUNC_PTR(TYPE, CN) \
1885 func = (ippiResizeFunc)ippiResizeCubic_##TYPE##_##CN##R; CHECK_FUNC(func);\
1886 status = ippiResizeGetSize_##TYPE(srcSize, dstSize, (IppiInterpolationType)mode, 0, &specSize, &initSize); CHECK_STATUS(status)\
1887 specBuf.allocate(specSize);\
1888 pSpec = (uchar*)specBuf;\
1889 status = ippiResizeCubicInit_##TYPE(srcSize, dstSize, 0.f, 0.75f, (IppiResizeSpec_32f*)pSpec, pInit); CHECK_STATUS(status);
1891 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR*100 + IPP_VERSION_MINOR >= 701)
1892 class IPPresizeInvoker :
1893 public ParallelLoopBody
1896 IPPresizeInvoker(Mat &_src, Mat &_dst, double _inv_scale_x, double _inv_scale_y, int _mode, bool *_ok) :
1897 ParallelLoopBody(), src(_src), dst(_dst), inv_scale_x(_inv_scale_x), inv_scale_y(_inv_scale_y), mode(_mode), ok(_ok)
1899 IppStatus status = ippStsNotSupportedModeErr;
1901 IppiSize srcSize, dstSize;
1902 int type = src.type();
1903 int specSize = 0, initSize = 0;
1904 srcSize.width = src.cols;
1905 srcSize.height = src.rows;
1906 dstSize.width = dst.cols;
1907 dstSize.height = dst.rows;
1909 if (mode == (int)ippLinear)
1913 case CV_8UC1: SET_IPP_RESIZE_LINEAR_FUNC_PTR(8u,C1); break;
1914 case CV_8UC3: SET_IPP_RESIZE_LINEAR_FUNC_PTR(8u,C3); break;
1915 case CV_8UC4: SET_IPP_RESIZE_LINEAR_FUNC_PTR(8u,C4); break;
1916 case CV_16UC1: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16u,C1); break;
1917 case CV_16UC3: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16u,C3); break;
1918 case CV_16UC4: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16u,C4); break;
1919 case CV_16SC1: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16s,C1); break;
1920 case CV_16SC3: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16s,C3); break;
1921 case CV_16SC4: SET_IPP_RESIZE_LINEAR_FUNC_PTR(16s,C4); break;
1922 case CV_32FC1: SET_IPP_RESIZE_LINEAR_FUNC_PTR(32f,C1); break;
1923 case CV_32FC3: SET_IPP_RESIZE_LINEAR_FUNC_PTR(32f,C3); break;
1924 case CV_32FC4: SET_IPP_RESIZE_LINEAR_FUNC_PTR(32f,C4); break;
1925 case CV_64FC1: SET_IPP_RESIZE_LINEAR_FUNC_64_PTR(64f,C1); break;
1926 case CV_64FC3: SET_IPP_RESIZE_LINEAR_FUNC_64_PTR(64f,C3); break;
1927 case CV_64FC4: SET_IPP_RESIZE_LINEAR_FUNC_64_PTR(64f,C4); break;
1928 default: { *ok = false; return;} break;
1931 else if (mode == (int)ippCubic)
1933 AutoBuffer<uchar> buf(initSize);
1934 uchar* pInit = (uchar*)buf;
1937 case CV_8UC1: SET_IPP_RESIZE_CUBIC_FUNC_PTR(8u,C1); break;
1938 case CV_8UC3: SET_IPP_RESIZE_CUBIC_FUNC_PTR(8u,C3); break;
1939 case CV_8UC4: SET_IPP_RESIZE_CUBIC_FUNC_PTR(8u,C4); break;
1940 case CV_16UC1: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16u,C1); break;
1941 case CV_16UC3: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16u,C3); break;
1942 case CV_16UC4: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16u,C4); break;
1943 case CV_16SC1: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16s,C1); break;
1944 case CV_16SC3: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16s,C3); break;
1945 case CV_16SC4: SET_IPP_RESIZE_CUBIC_FUNC_PTR(16s,C4); break;
1946 case CV_32FC1: SET_IPP_RESIZE_CUBIC_FUNC_PTR(32f,C1); break;
1947 case CV_32FC3: SET_IPP_RESIZE_CUBIC_FUNC_PTR(32f,C3); break;
1948 case CV_32FC4: SET_IPP_RESIZE_CUBIC_FUNC_PTR(32f,C4); break;
1949 default: { *ok = false; return;} break;
1958 virtual void operator() (const Range& range) const
1960 if (*ok == false) return;
1962 int cn = src.channels();
1963 int dsty = CV_IMIN(cvRound(range.start * inv_scale_y), dst.rows);
1964 int dstwidth = CV_IMIN(cvRound(src.cols * inv_scale_x), dst.cols);
1965 int dstheight = CV_IMIN(cvRound(range.end * inv_scale_y), dst.rows);
1967 IppiPoint dstOffset = { 0, dsty }, srcOffset = {0, 0};
1968 IppiSize dstSize = { dstwidth, dstheight - dsty };
1969 int bufsize = 0, itemSize = 0;
1971 IppStatus status = ippStsNotSupportedModeErr;
1973 switch (src.depth())
1977 status = ippiResizeGetBufferSize_8u((IppiResizeSpec_32f*)pSpec, dstSize, cn, &bufsize);
1978 CHECK_STATUS(status);
1979 status = ippiResizeGetSrcOffset_8u((IppiResizeSpec_32f*)pSpec, dstOffset, &srcOffset);
1983 status = ippiResizeGetBufferSize_16u((IppiResizeSpec_32f*)pSpec, dstSize, cn, &bufsize);
1984 CHECK_STATUS(status);
1985 status = ippiResizeGetSrcOffset_16u((IppiResizeSpec_32f*)pSpec, dstOffset, &srcOffset);
1989 status = ippiResizeGetBufferSize_16s((IppiResizeSpec_32f*)pSpec, dstSize, cn, &bufsize);
1990 CHECK_STATUS(status);
1991 status = ippiResizeGetSrcOffset_16s((IppiResizeSpec_32f*)pSpec, dstOffset, &srcOffset);
1995 status = ippiResizeGetBufferSize_32f((IppiResizeSpec_32f*)pSpec, dstSize, cn, &bufsize);
1996 CHECK_STATUS(status);
1997 status = ippiResizeGetSrcOffset_32f((IppiResizeSpec_32f*)pSpec, dstOffset, &srcOffset);
2001 status = ippiResizeGetBufferSize_64f((IppiResizeSpec_64f*)pSpec, dstSize, cn, &bufsize);
2002 CHECK_STATUS(status);
2003 status = ippiResizeGetSrcOffset_64f((IppiResizeSpec_64f*)pSpec, dstOffset, &srcOffset);
2007 CHECK_STATUS(status);
2009 Ipp8u* pSrc = (Ipp8u*)src.data + (int)src.step[0] * srcOffset.y + srcOffset.x * cn * itemSize;
2010 Ipp8u* pDst = (Ipp8u*)dst.data + (int)dst.step[0] * dstOffset.y + dstOffset.x * cn * itemSize;
2012 AutoBuffer<uchar> buf(bufsize + 64);
2013 uchar* bufptr = alignPtr((uchar*)buf, 32);
2017 if( func( pSrc, (int)src.step[0], pDst, (int)dst.step[0], dstOffset, dstSize, ippBorderRepl, 0, pSpec, bufptr ) < 0 )
2026 AutoBuffer<uchar> specBuf;
2028 ippiResizeFunc func;
2030 const IPPresizeInvoker& operator= (const IPPresizeInvoker&);
2036 static void ocl_computeResizeAreaTabs(int ssize, int dsize, double scale, int * const map_tab,
2037 float * const alpha_tab, int * const ofs_tab)
2040 for ( ; dx < dsize; dx++)
2044 double fsx1 = dx * scale;
2045 double fsx2 = fsx1 + scale;
2046 double cellWidth = std::min(scale, ssize - fsx1);
2048 int sx1 = cvCeil(fsx1), sx2 = cvFloor(fsx2);
2050 sx2 = std::min(sx2, ssize - 1);
2051 sx1 = std::min(sx1, sx2);
2053 if (sx1 - fsx1 > 1e-3)
2055 map_tab[k] = sx1 - 1;
2056 alpha_tab[k++] = (float)((sx1 - fsx1) / cellWidth);
2059 for (int sx = sx1; sx < sx2; sx++)
2062 alpha_tab[k++] = float(1.0 / cellWidth);
2065 if (fsx2 - sx2 > 1e-3)
2068 alpha_tab[k++] = (float)(std::min(std::min(fsx2 - sx2, 1.), cellWidth) / cellWidth);
2074 static void ocl_computeResizeAreaFastTabs(int * dmap_tab, int * smap_tab, int scale, int dcols, int scol)
2076 for (int i = 0; i < dcols; ++i)
2077 dmap_tab[i] = scale * i;
2079 for (int i = 0, size = dcols * scale; i < size; ++i)
2080 smap_tab[i] = std::min(scol - 1, i);
2083 static bool ocl_resize( InputArray _src, OutputArray _dst, Size dsize,
2084 double fx, double fy, int interpolation)
2086 int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
2088 double inv_fx = 1. / fx, inv_fy = 1. / fy;
2089 float inv_fxf = (float)inv_fx, inv_fyf = (float)inv_fy;
2092 (interpolation == INTER_NEAREST || interpolation == INTER_LINEAR ||
2093 (interpolation == INTER_AREA && inv_fx >= 1 && inv_fy >= 1) )) )
2096 UMat src = _src.getUMat();
2097 _dst.create(dsize, type);
2098 UMat dst = _dst.getUMat();
2101 size_t globalsize[] = { dst.cols, dst.rows };
2103 if (interpolation == INTER_LINEAR)
2105 int wdepth = std::max(depth, CV_32S);
2106 int wtype = CV_MAKETYPE(wdepth, cn);
2108 k.create("resizeLN", ocl::imgproc::resize_oclsrc,
2109 format("-D INTER_LINEAR -D depth=%d -D PIXTYPE=%s -D PIXTYPE1=%s "
2110 "-D WORKTYPE=%s -D convertToWT=%s -D convertToDT=%s -D cn=%d",
2111 depth, ocl::typeToStr(type), ocl::typeToStr(depth), ocl::typeToStr(wtype),
2112 ocl::convertTypeStr(depth, wdepth, cn, buf[0]),
2113 ocl::convertTypeStr(wdepth, depth, cn, buf[1]),
2116 else if (interpolation == INTER_NEAREST)
2118 k.create("resizeNN", ocl::imgproc::resize_oclsrc,
2119 format("-D INTER_NEAREST -D PIXTYPE=%s -D PIXTYPE1=%s -D cn=%d",
2120 ocl::memopTypeToStr(type), ocl::memopTypeToStr(depth), cn));
2122 else if (interpolation == INTER_AREA)
2124 int iscale_x = saturate_cast<int>(inv_fx);
2125 int iscale_y = saturate_cast<int>(inv_fy);
2126 bool is_area_fast = std::abs(inv_fx - iscale_x) < DBL_EPSILON &&
2127 std::abs(inv_fy - iscale_y) < DBL_EPSILON;
2128 int wdepth = std::max(depth, is_area_fast ? CV_32S : CV_32F);
2129 int wtype = CV_MAKE_TYPE(wdepth, cn);
2132 String buildOption = format("-D INTER_AREA -D PIXTYPE=%s -D PIXTYPE1=%s -D WTV=%s -D convertToWTV=%s -D cn=%d",
2133 ocl::typeToStr(type), ocl::typeToStr(depth), ocl::typeToStr(wtype),
2134 ocl::convertTypeStr(depth, wdepth, cn, cvt[0]), cn);
2136 UMat alphaOcl, tabofsOcl, mapOcl;
2141 int wdepth2 = std::max(CV_32F, depth), wtype2 = CV_MAKE_TYPE(wdepth2, cn);
2142 buildOption = buildOption + format(" -D convertToPIXTYPE=%s -D WT2V=%s -D convertToWT2V=%s -D INTER_AREA_FAST"
2143 " -D XSCALE=%d -D YSCALE=%d -D SCALE=%ff",
2144 ocl::convertTypeStr(wdepth2, depth, cn, cvt[0]),
2145 ocl::typeToStr(wtype2), ocl::convertTypeStr(wdepth, wdepth2, cn, cvt[1]),
2146 iscale_x, iscale_y, 1.0f / (iscale_x * iscale_y));
2148 k.create("resizeAREA_FAST", ocl::imgproc::resize_oclsrc, buildOption);
2152 int smap_tab_size = dst.cols * iscale_x + dst.rows * iscale_y;
2153 AutoBuffer<int> dmap_tab(dst.cols + dst.rows), smap_tab(smap_tab_size);
2154 int * dxmap_tab = dmap_tab, * dymap_tab = dxmap_tab + dst.cols;
2155 int * sxmap_tab = smap_tab, * symap_tab = smap_tab + dst.cols * iscale_y;
2157 ocl_computeResizeAreaFastTabs(dxmap_tab, sxmap_tab, iscale_x, dst.cols, src.cols);
2158 ocl_computeResizeAreaFastTabs(dymap_tab, symap_tab, iscale_y, dst.rows, src.rows);
2160 Mat(1, dst.cols + dst.rows, CV_32SC1, (void *)dmap_tab).copyTo(dmap);
2161 Mat(1, smap_tab_size, CV_32SC1, (void *)smap_tab).copyTo(smap);
2165 buildOption = buildOption + format(" -D convertToPIXTYPE=%s", ocl::convertTypeStr(wdepth, depth, cn, cvt[0]));
2166 k.create("resizeAREA", ocl::imgproc::resize_oclsrc, buildOption);
2170 Size ssize = src.size();
2171 int xytab_size = (ssize.width + ssize.height) << 1;
2172 int tabofs_size = dsize.height + dsize.width + 2;
2174 AutoBuffer<int> _xymap_tab(xytab_size), _xyofs_tab(tabofs_size);
2175 AutoBuffer<float> _xyalpha_tab(xytab_size);
2176 int * xmap_tab = _xymap_tab, * ymap_tab = _xymap_tab + (ssize.width << 1);
2177 float * xalpha_tab = _xyalpha_tab, * yalpha_tab = _xyalpha_tab + (ssize.width << 1);
2178 int * xofs_tab = _xyofs_tab, * yofs_tab = _xyofs_tab + dsize.width + 1;
2180 ocl_computeResizeAreaTabs(ssize.width, dsize.width, inv_fx, xmap_tab, xalpha_tab, xofs_tab);
2181 ocl_computeResizeAreaTabs(ssize.height, dsize.height, inv_fy, ymap_tab, yalpha_tab, yofs_tab);
2183 // loading precomputed arrays to GPU
2184 Mat(1, xytab_size, CV_32FC1, (void *)_xyalpha_tab).copyTo(alphaOcl);
2185 Mat(1, xytab_size, CV_32SC1, (void *)_xymap_tab).copyTo(mapOcl);
2186 Mat(1, tabofs_size, CV_32SC1, (void *)_xyofs_tab).copyTo(tabofsOcl);
2189 ocl::KernelArg srcarg = ocl::KernelArg::ReadOnly(src), dstarg = ocl::KernelArg::WriteOnly(dst);
2192 k.args(srcarg, dstarg, ocl::KernelArg::PtrReadOnly(dmap), ocl::KernelArg::PtrReadOnly(smap));
2194 k.args(srcarg, dstarg, inv_fxf, inv_fyf, ocl::KernelArg::PtrReadOnly(tabofsOcl),
2195 ocl::KernelArg::PtrReadOnly(mapOcl), ocl::KernelArg::PtrReadOnly(alphaOcl));
2197 return k.run(2, globalsize, NULL, false);
2202 k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst),
2203 (float)inv_fx, (float)inv_fy);
2205 return k.run(2, globalsize, 0, false);
2212 //////////////////////////////////////////////////////////////////////////////////////////
2214 void cv::resize( InputArray _src, OutputArray _dst, Size dsize,
2215 double inv_scale_x, double inv_scale_y, int interpolation )
2217 static ResizeFunc linear_tab[] =
2220 HResizeLinear<uchar, int, short,
2221 INTER_RESIZE_COEF_SCALE,
2222 HResizeLinearVec_8u32s>,
2223 VResizeLinear<uchar, int, short,
2224 FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS*2>,
2225 VResizeLinearVec_32s8u> >,
2228 HResizeLinear<ushort, float, float, 1,
2229 HResizeLinearVec_16u32f>,
2230 VResizeLinear<ushort, float, float, Cast<float, ushort>,
2231 VResizeLinearVec_32f16u> >,
2233 HResizeLinear<short, float, float, 1,
2234 HResizeLinearVec_16s32f>,
2235 VResizeLinear<short, float, float, Cast<float, short>,
2236 VResizeLinearVec_32f16s> >,
2239 HResizeLinear<float, float, float, 1,
2240 HResizeLinearVec_32f>,
2241 VResizeLinear<float, float, float, Cast<float, float>,
2242 VResizeLinearVec_32f> >,
2244 HResizeLinear<double, double, float, 1,
2246 VResizeLinear<double, double, float, Cast<double, double>,
2251 static ResizeFunc cubic_tab[] =
2254 HResizeCubic<uchar, int, short>,
2255 VResizeCubic<uchar, int, short,
2256 FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS*2>,
2257 VResizeCubicVec_32s8u> >,
2260 HResizeCubic<ushort, float, float>,
2261 VResizeCubic<ushort, float, float, Cast<float, ushort>,
2262 VResizeCubicVec_32f16u> >,
2264 HResizeCubic<short, float, float>,
2265 VResizeCubic<short, float, float, Cast<float, short>,
2266 VResizeCubicVec_32f16s> >,
2269 HResizeCubic<float, float, float>,
2270 VResizeCubic<float, float, float, Cast<float, float>,
2271 VResizeCubicVec_32f> >,
2273 HResizeCubic<double, double, float>,
2274 VResizeCubic<double, double, float, Cast<double, double>,
2279 static ResizeFunc lanczos4_tab[] =
2281 resizeGeneric_<HResizeLanczos4<uchar, int, short>,
2282 VResizeLanczos4<uchar, int, short,
2283 FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS*2>,
2286 resizeGeneric_<HResizeLanczos4<ushort, float, float>,
2287 VResizeLanczos4<ushort, float, float, Cast<float, ushort>,
2289 resizeGeneric_<HResizeLanczos4<short, float, float>,
2290 VResizeLanczos4<short, float, float, Cast<float, short>,
2293 resizeGeneric_<HResizeLanczos4<float, float, float>,
2294 VResizeLanczos4<float, float, float, Cast<float, float>,
2296 resizeGeneric_<HResizeLanczos4<double, double, float>,
2297 VResizeLanczos4<double, double, float, Cast<double, double>,
2302 static ResizeAreaFastFunc areafast_tab[] =
2304 resizeAreaFast_<uchar, int, ResizeAreaFastVec<uchar, ResizeAreaFastVec_SIMD_8u> >,
2306 resizeAreaFast_<ushort, float, ResizeAreaFastVec<ushort, ResizeAreaFastVec_SIMD_16u> >,
2307 resizeAreaFast_<short, float, ResizeAreaFastVec<short, ResizeAreaFastNoVec<short, float> > >,
2309 resizeAreaFast_<float, float, ResizeAreaFastNoVec<float, float> >,
2310 resizeAreaFast_<double, double, ResizeAreaFastNoVec<double, double> >,
2314 static ResizeAreaFunc area_tab[] =
2316 resizeArea_<uchar, float>, 0, resizeArea_<ushort, float>,
2317 resizeArea_<short, float>, 0, resizeArea_<float, float>,
2318 resizeArea_<double, double>, 0
2321 Size ssize = _src.size();
2323 CV_Assert( ssize.area() > 0 );
2324 CV_Assert( dsize.area() > 0 || (inv_scale_x > 0 && inv_scale_y > 0) );
2325 if( dsize.area() == 0 )
2327 dsize = Size(saturate_cast<int>(ssize.width*inv_scale_x),
2328 saturate_cast<int>(ssize.height*inv_scale_y));
2329 CV_Assert( dsize.area() > 0 );
2333 inv_scale_x = (double)dsize.width/ssize.width;
2334 inv_scale_y = (double)dsize.height/ssize.height;
2337 CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
2338 ocl_resize(_src, _dst, dsize, inv_scale_x, inv_scale_y, interpolation))
2340 Mat src = _src.getMat();
2341 _dst.create(dsize, src.type());
2342 Mat dst = _dst.getMat();
2344 #ifdef HAVE_TEGRA_OPTIMIZATION
2345 if (tegra::resize(src, dst, (float)inv_scale_x, (float)inv_scale_y, interpolation))
2349 int depth = src.depth(), cn = src.channels();
2350 double scale_x = 1./inv_scale_x, scale_y = 1./inv_scale_y;
2351 int k, sx, sy, dx, dy;
2353 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR*100 + IPP_VERSION_MINOR >= 701)
2354 #define IPP_RESIZE_EPS 1.e-10
2356 double ex = fabs((double)dsize.width/src.cols - inv_scale_x)/inv_scale_x;
2357 double ey = fabs((double)dsize.height/src.rows - inv_scale_y)/inv_scale_y;
2359 if ((ex < IPP_RESIZE_EPS && ey < IPP_RESIZE_EPS && depth != CV_64F) ||
2360 (ex == 0 && ey == 0 && depth == CV_64F))
2363 if (interpolation == INTER_LINEAR && src.rows >= 2 && src.cols >= 2)
2367 else if (interpolation == INTER_CUBIC && src.rows >= 4 && src.cols >= 4)
2371 if( mode != 0 && (cn == 1 || cn ==3 || cn == 4) &&
2372 (depth == CV_8U || depth == CV_16U || depth == CV_16S || depth == CV_32F ||
2373 (depth == CV_64F && mode == ippLinear)))
2376 Range range(0, src.rows);
2377 IPPresizeInvoker invoker(src, dst, inv_scale_x, inv_scale_y, mode, &ok);
2378 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
2383 #undef IPP_RESIZE_EPS
2386 if( interpolation == INTER_NEAREST )
2388 resizeNN( src, dst, inv_scale_x, inv_scale_y );
2393 int iscale_x = saturate_cast<int>(scale_x);
2394 int iscale_y = saturate_cast<int>(scale_y);
2396 bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON &&
2397 std::abs(scale_y - iscale_y) < DBL_EPSILON;
2399 // in case of scale_x && scale_y is equal to 2
2400 // INTER_AREA (fast) also is equal to INTER_LINEAR
2401 if( interpolation == INTER_LINEAR && is_area_fast && iscale_x == 2 && iscale_y == 2 )
2402 interpolation = INTER_AREA;
2404 // true "area" interpolation is only implemented for the case (scale_x <= 1 && scale_y <= 1).
2405 // In other cases it is emulated using some variant of bilinear interpolation
2406 if( interpolation == INTER_AREA && scale_x >= 1 && scale_y >= 1 )
2410 int area = iscale_x*iscale_y;
2411 size_t srcstep = src.step / src.elemSize1();
2412 AutoBuffer<int> _ofs(area + dsize.width*cn);
2414 int* xofs = ofs + area;
2415 ResizeAreaFastFunc func = areafast_tab[depth];
2416 CV_Assert( func != 0 );
2418 for( sy = 0, k = 0; sy < iscale_y; sy++ )
2419 for( sx = 0; sx < iscale_x; sx++ )
2420 ofs[k++] = (int)(sy*srcstep + sx*cn);
2422 for( dx = 0; dx < dsize.width; dx++ )
2426 for( k = 0; k < cn; k++ )
2427 xofs[j + k] = sx + k;
2430 func( src, dst, ofs, xofs, iscale_x, iscale_y );
2434 ResizeAreaFunc func = area_tab[depth];
2435 CV_Assert( func != 0 && cn <= 4 );
2437 AutoBuffer<DecimateAlpha> _xytab((ssize.width + ssize.height)*2);
2438 DecimateAlpha* xtab = _xytab, *ytab = xtab + ssize.width*2;
2440 int xtab_size = computeResizeAreaTab(ssize.width, dsize.width, cn, scale_x, xtab);
2441 int ytab_size = computeResizeAreaTab(ssize.height, dsize.height, 1, scale_y, ytab);
2443 AutoBuffer<int> _tabofs(dsize.height + 1);
2444 int* tabofs = _tabofs;
2445 for( k = 0, dy = 0; k < ytab_size; k++ )
2447 if( k == 0 || ytab[k].di != ytab[k-1].di )
2449 assert( ytab[k].di == dy );
2453 tabofs[dy] = ytab_size;
2455 func( src, dst, xtab, xtab_size, ytab, ytab_size, tabofs );
2460 int xmin = 0, xmax = dsize.width, width = dsize.width*cn;
2461 bool area_mode = interpolation == INTER_AREA;
2462 bool fixpt = depth == CV_8U;
2465 int ksize=0, ksize2;
2466 if( interpolation == INTER_CUBIC )
2467 ksize = 4, func = cubic_tab[depth];
2468 else if( interpolation == INTER_LANCZOS4 )
2469 ksize = 8, func = lanczos4_tab[depth];
2470 else if( interpolation == INTER_LINEAR || interpolation == INTER_AREA )
2471 ksize = 2, func = linear_tab[depth];
2473 CV_Error( CV_StsBadArg, "Unknown interpolation method" );
2476 CV_Assert( func != 0 );
2478 AutoBuffer<uchar> _buffer((width + dsize.height)*(sizeof(int) + sizeof(float)*ksize));
2479 int* xofs = (int*)(uchar*)_buffer;
2480 int* yofs = xofs + width;
2481 float* alpha = (float*)(yofs + dsize.height);
2482 short* ialpha = (short*)alpha;
2483 float* beta = alpha + width*ksize;
2484 short* ibeta = ialpha + width*ksize;
2485 float cbuf[MAX_ESIZE];
2487 for( dx = 0; dx < dsize.width; dx++ )
2491 fx = (float)((dx+0.5)*scale_x - 0.5);
2497 sx = cvFloor(dx*scale_x);
2498 fx = (float)((dx+1) - (sx+1)*inv_scale_x);
2499 fx = fx <= 0 ? 0.f : fx - cvFloor(fx);
2505 if( sx < 0 && (interpolation != INTER_CUBIC && interpolation != INTER_LANCZOS4))
2509 if( sx + ksize2 >= ssize.width )
2511 xmax = std::min( xmax, dx );
2512 if( sx >= ssize.width-1 && (interpolation != INTER_CUBIC && interpolation != INTER_LANCZOS4))
2513 fx = 0, sx = ssize.width-1;
2516 for( k = 0, sx *= cn; k < cn; k++ )
2517 xofs[dx*cn + k] = sx + k;
2519 if( interpolation == INTER_CUBIC )
2520 interpolateCubic( fx, cbuf );
2521 else if( interpolation == INTER_LANCZOS4 )
2522 interpolateLanczos4( fx, cbuf );
2530 for( k = 0; k < ksize; k++ )
2531 ialpha[dx*cn*ksize + k] = saturate_cast<short>(cbuf[k]*INTER_RESIZE_COEF_SCALE);
2532 for( ; k < cn*ksize; k++ )
2533 ialpha[dx*cn*ksize + k] = ialpha[dx*cn*ksize + k - ksize];
2537 for( k = 0; k < ksize; k++ )
2538 alpha[dx*cn*ksize + k] = cbuf[k];
2539 for( ; k < cn*ksize; k++ )
2540 alpha[dx*cn*ksize + k] = alpha[dx*cn*ksize + k - ksize];
2544 for( dy = 0; dy < dsize.height; dy++ )
2548 fy = (float)((dy+0.5)*scale_y - 0.5);
2554 sy = cvFloor(dy*scale_y);
2555 fy = (float)((dy+1) - (sy+1)*inv_scale_y);
2556 fy = fy <= 0 ? 0.f : fy - cvFloor(fy);
2560 if( interpolation == INTER_CUBIC )
2561 interpolateCubic( fy, cbuf );
2562 else if( interpolation == INTER_LANCZOS4 )
2563 interpolateLanczos4( fy, cbuf );
2572 for( k = 0; k < ksize; k++ )
2573 ibeta[dy*ksize + k] = saturate_cast<short>(cbuf[k]*INTER_RESIZE_COEF_SCALE);
2577 for( k = 0; k < ksize; k++ )
2578 beta[dy*ksize + k] = cbuf[k];
2582 func( src, dst, xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs,
2583 fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize );
2587 /****************************************************************************************\
2588 * General warping (affine, perspective, remap) *
2589 \****************************************************************************************/
2594 template<typename T>
2595 static void remapNearest( const Mat& _src, Mat& _dst, const Mat& _xy,
2596 int borderType, const Scalar& _borderValue )
2598 Size ssize = _src.size(), dsize = _dst.size();
2599 int cn = _src.channels();
2600 const T* S0 = (const T*)_src.data;
2601 size_t sstep = _src.step/sizeof(S0[0]);
2602 Scalar_<T> cval(saturate_cast<T>(_borderValue[0]),
2603 saturate_cast<T>(_borderValue[1]),
2604 saturate_cast<T>(_borderValue[2]),
2605 saturate_cast<T>(_borderValue[3]));
2608 unsigned width1 = ssize.width, height1 = ssize.height;
2610 if( _dst.isContinuous() && _xy.isContinuous() )
2612 dsize.width *= dsize.height;
2616 for( dy = 0; dy < dsize.height; dy++ )
2618 T* D = (T*)(_dst.data + _dst.step*dy);
2619 const short* XY = (const short*)(_xy.data + _xy.step*dy);
2623 for( dx = 0; dx < dsize.width; dx++ )
2625 int sx = XY[dx*2], sy = XY[dx*2+1];
2626 if( (unsigned)sx < width1 && (unsigned)sy < height1 )
2627 D[dx] = S0[sy*sstep + sx];
2630 if( borderType == BORDER_REPLICATE )
2632 sx = clip(sx, 0, ssize.width);
2633 sy = clip(sy, 0, ssize.height);
2634 D[dx] = S0[sy*sstep + sx];
2636 else if( borderType == BORDER_CONSTANT )
2638 else if( borderType != BORDER_TRANSPARENT )
2640 sx = borderInterpolate(sx, ssize.width, borderType);
2641 sy = borderInterpolate(sy, ssize.height, borderType);
2642 D[dx] = S0[sy*sstep + sx];
2649 for( dx = 0; dx < dsize.width; dx++, D += cn )
2651 int sx = XY[dx*2], sy = XY[dx*2+1], k;
2653 if( (unsigned)sx < width1 && (unsigned)sy < height1 )
2657 S = S0 + sy*sstep + sx*3;
2658 D[0] = S[0], D[1] = S[1], D[2] = S[2];
2662 S = S0 + sy*sstep + sx*4;
2663 D[0] = S[0], D[1] = S[1], D[2] = S[2], D[3] = S[3];
2667 S = S0 + sy*sstep + sx*cn;
2668 for( k = 0; k < cn; k++ )
2672 else if( borderType != BORDER_TRANSPARENT )
2674 if( borderType == BORDER_REPLICATE )
2676 sx = clip(sx, 0, ssize.width);
2677 sy = clip(sy, 0, ssize.height);
2678 S = S0 + sy*sstep + sx*cn;
2680 else if( borderType == BORDER_CONSTANT )
2684 sx = borderInterpolate(sx, ssize.width, borderType);
2685 sy = borderInterpolate(sy, ssize.height, borderType);
2686 S = S0 + sy*sstep + sx*cn;
2688 for( k = 0; k < cn; k++ )
2699 int operator()( const Mat&, void*, const short*, const ushort*,
2700 const void*, int ) const { return 0; }
2707 int operator()( const Mat& _src, void* _dst, const short* XY,
2708 const ushort* FXY, const void* _wtab, int width ) const
2710 int cn = _src.channels(), x = 0, sstep = (int)_src.step;
2712 if( (cn != 1 && cn != 3 && cn != 4) || !checkHardwareSupport(CV_CPU_SSE2) ||
2716 const uchar *S0 = _src.data, *S1 = _src.data + _src.step;
2717 const short* wtab = cn == 1 ? (const short*)_wtab : &BilinearTab_iC4[0][0][0];
2718 uchar* D = (uchar*)_dst;
2719 __m128i delta = _mm_set1_epi32(INTER_REMAP_COEF_SCALE/2);
2720 __m128i xy2ofs = _mm_set1_epi32(cn + (sstep << 16));
2721 __m128i z = _mm_setzero_si128();
2722 int CV_DECL_ALIGNED(16) iofs0[4], iofs1[4];
2726 for( ; x <= width - 8; x += 8 )
2728 __m128i xy0 = _mm_loadu_si128( (const __m128i*)(XY + x*2));
2729 __m128i xy1 = _mm_loadu_si128( (const __m128i*)(XY + x*2 + 8));
2730 __m128i v0, v1, v2, v3, a0, a1, b0, b1;
2733 xy0 = _mm_madd_epi16( xy0, xy2ofs );
2734 xy1 = _mm_madd_epi16( xy1, xy2ofs );
2735 _mm_store_si128( (__m128i*)iofs0, xy0 );
2736 _mm_store_si128( (__m128i*)iofs1, xy1 );
2738 i0 = *(ushort*)(S0 + iofs0[0]) + (*(ushort*)(S0 + iofs0[1]) << 16);
2739 i1 = *(ushort*)(S0 + iofs0[2]) + (*(ushort*)(S0 + iofs0[3]) << 16);
2740 v0 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
2741 i0 = *(ushort*)(S1 + iofs0[0]) + (*(ushort*)(S1 + iofs0[1]) << 16);
2742 i1 = *(ushort*)(S1 + iofs0[2]) + (*(ushort*)(S1 + iofs0[3]) << 16);
2743 v1 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
2744 v0 = _mm_unpacklo_epi8(v0, z);
2745 v1 = _mm_unpacklo_epi8(v1, z);
2747 a0 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)(wtab+FXY[x]*4)),
2748 _mm_loadl_epi64((__m128i*)(wtab+FXY[x+1]*4)));
2749 a1 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)(wtab+FXY[x+2]*4)),
2750 _mm_loadl_epi64((__m128i*)(wtab+FXY[x+3]*4)));
2751 b0 = _mm_unpacklo_epi64(a0, a1);
2752 b1 = _mm_unpackhi_epi64(a0, a1);
2753 v0 = _mm_madd_epi16(v0, b0);
2754 v1 = _mm_madd_epi16(v1, b1);
2755 v0 = _mm_add_epi32(_mm_add_epi32(v0, v1), delta);
2757 i0 = *(ushort*)(S0 + iofs1[0]) + (*(ushort*)(S0 + iofs1[1]) << 16);
2758 i1 = *(ushort*)(S0 + iofs1[2]) + (*(ushort*)(S0 + iofs1[3]) << 16);
2759 v2 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
2760 i0 = *(ushort*)(S1 + iofs1[0]) + (*(ushort*)(S1 + iofs1[1]) << 16);
2761 i1 = *(ushort*)(S1 + iofs1[2]) + (*(ushort*)(S1 + iofs1[3]) << 16);
2762 v3 = _mm_unpacklo_epi32(_mm_cvtsi32_si128(i0), _mm_cvtsi32_si128(i1));
2763 v2 = _mm_unpacklo_epi8(v2, z);
2764 v3 = _mm_unpacklo_epi8(v3, z);
2766 a0 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)(wtab+FXY[x+4]*4)),
2767 _mm_loadl_epi64((__m128i*)(wtab+FXY[x+5]*4)));
2768 a1 = _mm_unpacklo_epi32(_mm_loadl_epi64((__m128i*)(wtab+FXY[x+6]*4)),
2769 _mm_loadl_epi64((__m128i*)(wtab+FXY[x+7]*4)));
2770 b0 = _mm_unpacklo_epi64(a0, a1);
2771 b1 = _mm_unpackhi_epi64(a0, a1);
2772 v2 = _mm_madd_epi16(v2, b0);
2773 v3 = _mm_madd_epi16(v3, b1);
2774 v2 = _mm_add_epi32(_mm_add_epi32(v2, v3), delta);
2776 v0 = _mm_srai_epi32(v0, INTER_REMAP_COEF_BITS);
2777 v2 = _mm_srai_epi32(v2, INTER_REMAP_COEF_BITS);
2778 v0 = _mm_packus_epi16(_mm_packs_epi32(v0, v2), z);
2779 _mm_storel_epi64( (__m128i*)(D + x), v0 );
2784 for( ; x <= width - 5; x += 4, D += 12 )
2786 __m128i xy0 = _mm_loadu_si128( (const __m128i*)(XY + x*2));
2787 __m128i u0, v0, u1, v1;
2789 xy0 = _mm_madd_epi16( xy0, xy2ofs );
2790 _mm_store_si128( (__m128i*)iofs0, xy0 );
2791 const __m128i *w0, *w1;
2792 w0 = (const __m128i*)(wtab + FXY[x]*16);
2793 w1 = (const __m128i*)(wtab + FXY[x+1]*16);
2795 u0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[0])),
2796 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[0] + 3)));
2797 v0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[0])),
2798 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[0] + 3)));
2799 u1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[1])),
2800 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[1] + 3)));
2801 v1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[1])),
2802 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[1] + 3)));
2803 u0 = _mm_unpacklo_epi8(u0, z);
2804 v0 = _mm_unpacklo_epi8(v0, z);
2805 u1 = _mm_unpacklo_epi8(u1, z);
2806 v1 = _mm_unpacklo_epi8(v1, z);
2807 u0 = _mm_add_epi32(_mm_madd_epi16(u0, w0[0]), _mm_madd_epi16(v0, w0[1]));
2808 u1 = _mm_add_epi32(_mm_madd_epi16(u1, w1[0]), _mm_madd_epi16(v1, w1[1]));
2809 u0 = _mm_srai_epi32(_mm_add_epi32(u0, delta), INTER_REMAP_COEF_BITS);
2810 u1 = _mm_srai_epi32(_mm_add_epi32(u1, delta), INTER_REMAP_COEF_BITS);
2811 u0 = _mm_slli_si128(u0, 4);
2812 u0 = _mm_packs_epi32(u0, u1);
2813 u0 = _mm_packus_epi16(u0, u0);
2814 _mm_storel_epi64((__m128i*)D, _mm_srli_si128(u0,1));
2816 w0 = (const __m128i*)(wtab + FXY[x+2]*16);
2817 w1 = (const __m128i*)(wtab + FXY[x+3]*16);
2819 u0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[2])),
2820 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[2] + 3)));
2821 v0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[2])),
2822 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[2] + 3)));
2823 u1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[3])),
2824 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[3] + 3)));
2825 v1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[3])),
2826 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[3] + 3)));
2827 u0 = _mm_unpacklo_epi8(u0, z);
2828 v0 = _mm_unpacklo_epi8(v0, z);
2829 u1 = _mm_unpacklo_epi8(u1, z);
2830 v1 = _mm_unpacklo_epi8(v1, z);
2831 u0 = _mm_add_epi32(_mm_madd_epi16(u0, w0[0]), _mm_madd_epi16(v0, w0[1]));
2832 u1 = _mm_add_epi32(_mm_madd_epi16(u1, w1[0]), _mm_madd_epi16(v1, w1[1]));
2833 u0 = _mm_srai_epi32(_mm_add_epi32(u0, delta), INTER_REMAP_COEF_BITS);
2834 u1 = _mm_srai_epi32(_mm_add_epi32(u1, delta), INTER_REMAP_COEF_BITS);
2835 u0 = _mm_slli_si128(u0, 4);
2836 u0 = _mm_packs_epi32(u0, u1);
2837 u0 = _mm_packus_epi16(u0, u0);
2838 _mm_storel_epi64((__m128i*)(D + 6), _mm_srli_si128(u0,1));
2843 for( ; x <= width - 4; x += 4, D += 16 )
2845 __m128i xy0 = _mm_loadu_si128( (const __m128i*)(XY + x*2));
2846 __m128i u0, v0, u1, v1;
2848 xy0 = _mm_madd_epi16( xy0, xy2ofs );
2849 _mm_store_si128( (__m128i*)iofs0, xy0 );
2850 const __m128i *w0, *w1;
2851 w0 = (const __m128i*)(wtab + FXY[x]*16);
2852 w1 = (const __m128i*)(wtab + FXY[x+1]*16);
2854 u0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[0])),
2855 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[0] + 4)));
2856 v0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[0])),
2857 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[0] + 4)));
2858 u1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[1])),
2859 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[1] + 4)));
2860 v1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[1])),
2861 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[1] + 4)));
2862 u0 = _mm_unpacklo_epi8(u0, z);
2863 v0 = _mm_unpacklo_epi8(v0, z);
2864 u1 = _mm_unpacklo_epi8(u1, z);
2865 v1 = _mm_unpacklo_epi8(v1, z);
2866 u0 = _mm_add_epi32(_mm_madd_epi16(u0, w0[0]), _mm_madd_epi16(v0, w0[1]));
2867 u1 = _mm_add_epi32(_mm_madd_epi16(u1, w1[0]), _mm_madd_epi16(v1, w1[1]));
2868 u0 = _mm_srai_epi32(_mm_add_epi32(u0, delta), INTER_REMAP_COEF_BITS);
2869 u1 = _mm_srai_epi32(_mm_add_epi32(u1, delta), INTER_REMAP_COEF_BITS);
2870 u0 = _mm_packs_epi32(u0, u1);
2871 u0 = _mm_packus_epi16(u0, u0);
2872 _mm_storel_epi64((__m128i*)D, u0);
2874 w0 = (const __m128i*)(wtab + FXY[x+2]*16);
2875 w1 = (const __m128i*)(wtab + FXY[x+3]*16);
2877 u0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[2])),
2878 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[2] + 4)));
2879 v0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[2])),
2880 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[2] + 4)));
2881 u1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S0 + iofs0[3])),
2882 _mm_cvtsi32_si128(*(int*)(S0 + iofs0[3] + 4)));
2883 v1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(int*)(S1 + iofs0[3])),
2884 _mm_cvtsi32_si128(*(int*)(S1 + iofs0[3] + 4)));
2885 u0 = _mm_unpacklo_epi8(u0, z);
2886 v0 = _mm_unpacklo_epi8(v0, z);
2887 u1 = _mm_unpacklo_epi8(u1, z);
2888 v1 = _mm_unpacklo_epi8(v1, z);
2889 u0 = _mm_add_epi32(_mm_madd_epi16(u0, w0[0]), _mm_madd_epi16(v0, w0[1]));
2890 u1 = _mm_add_epi32(_mm_madd_epi16(u1, w1[0]), _mm_madd_epi16(v1, w1[1]));
2891 u0 = _mm_srai_epi32(_mm_add_epi32(u0, delta), INTER_REMAP_COEF_BITS);
2892 u1 = _mm_srai_epi32(_mm_add_epi32(u1, delta), INTER_REMAP_COEF_BITS);
2893 u0 = _mm_packs_epi32(u0, u1);
2894 u0 = _mm_packus_epi16(u0, u0);
2895 _mm_storel_epi64((__m128i*)(D + 8), u0);
2905 typedef RemapNoVec RemapVec_8u;
2910 template<class CastOp, class VecOp, typename AT>
2911 static void remapBilinear( const Mat& _src, Mat& _dst, const Mat& _xy,
2912 const Mat& _fxy, const void* _wtab,
2913 int borderType, const Scalar& _borderValue )
2915 typedef typename CastOp::rtype T;
2916 typedef typename CastOp::type1 WT;
2917 Size ssize = _src.size(), dsize = _dst.size();
2918 int cn = _src.channels();
2919 const AT* wtab = (const AT*)_wtab;
2920 const T* S0 = (const T*)_src.data;
2921 size_t sstep = _src.step/sizeof(S0[0]);
2922 Scalar_<T> cval(saturate_cast<T>(_borderValue[0]),
2923 saturate_cast<T>(_borderValue[1]),
2924 saturate_cast<T>(_borderValue[2]),
2925 saturate_cast<T>(_borderValue[3]));
2930 unsigned width1 = std::max(ssize.width-1, 0), height1 = std::max(ssize.height-1, 0);
2931 CV_Assert( cn <= 4 && ssize.area() > 0 );
2933 if( _src.type() == CV_8UC3 )
2934 width1 = std::max(ssize.width-2, 0);
2937 for( dy = 0; dy < dsize.height; dy++ )
2939 T* D = (T*)(_dst.data + _dst.step*dy);
2940 const short* XY = (const short*)(_xy.data + _xy.step*dy);
2941 const ushort* FXY = (const ushort*)(_fxy.data + _fxy.step*dy);
2943 bool prevInlier = false;
2945 for( dx = 0; dx <= dsize.width; dx++ )
2947 bool curInlier = dx < dsize.width ?
2948 (unsigned)XY[dx*2] < width1 &&
2949 (unsigned)XY[dx*2+1] < height1 : !prevInlier;
2950 if( curInlier == prevInlier )
2956 prevInlier = curInlier;
2960 int len = vecOp( _src, D, XY + dx*2, FXY + dx, wtab, X1 - dx );
2966 for( ; dx < X1; dx++, D++ )
2968 int sx = XY[dx*2], sy = XY[dx*2+1];
2969 const AT* w = wtab + FXY[dx]*4;
2970 const T* S = S0 + sy*sstep + sx;
2971 *D = castOp(WT(S[0]*w[0] + S[1]*w[1] + S[sstep]*w[2] + S[sstep+1]*w[3]));
2975 for( ; dx < X1; dx++, D += 2 )
2977 int sx = XY[dx*2], sy = XY[dx*2+1];
2978 const AT* w = wtab + FXY[dx]*4;
2979 const T* S = S0 + sy*sstep + sx*2;
2980 WT t0 = S[0]*w[0] + S[2]*w[1] + S[sstep]*w[2] + S[sstep+2]*w[3];
2981 WT t1 = S[1]*w[0] + S[3]*w[1] + S[sstep+1]*w[2] + S[sstep+3]*w[3];
2982 D[0] = castOp(t0); D[1] = castOp(t1);
2985 for( ; dx < X1; dx++, D += 3 )
2987 int sx = XY[dx*2], sy = XY[dx*2+1];
2988 const AT* w = wtab + FXY[dx]*4;
2989 const T* S = S0 + sy*sstep + sx*3;
2990 WT t0 = S[0]*w[0] + S[3]*w[1] + S[sstep]*w[2] + S[sstep+3]*w[3];
2991 WT t1 = S[1]*w[0] + S[4]*w[1] + S[sstep+1]*w[2] + S[sstep+4]*w[3];
2992 WT t2 = S[2]*w[0] + S[5]*w[1] + S[sstep+2]*w[2] + S[sstep+5]*w[3];
2993 D[0] = castOp(t0); D[1] = castOp(t1); D[2] = castOp(t2);
2996 for( ; dx < X1; dx++, D += 4 )
2998 int sx = XY[dx*2], sy = XY[dx*2+1];
2999 const AT* w = wtab + FXY[dx]*4;
3000 const T* S = S0 + sy*sstep + sx*4;
3001 WT t0 = S[0]*w[0] + S[4]*w[1] + S[sstep]*w[2] + S[sstep+4]*w[3];
3002 WT t1 = S[1]*w[0] + S[5]*w[1] + S[sstep+1]*w[2] + S[sstep+5]*w[3];
3003 D[0] = castOp(t0); D[1] = castOp(t1);
3004 t0 = S[2]*w[0] + S[6]*w[1] + S[sstep+2]*w[2] + S[sstep+6]*w[3];
3005 t1 = S[3]*w[0] + S[7]*w[1] + S[sstep+3]*w[2] + S[sstep+7]*w[3];
3006 D[2] = castOp(t0); D[3] = castOp(t1);
3011 if( borderType == BORDER_TRANSPARENT && cn != 3 )
3019 for( ; dx < X1; dx++, D++ )
3021 int sx = XY[dx*2], sy = XY[dx*2+1];
3022 if( borderType == BORDER_CONSTANT &&
3023 (sx >= ssize.width || sx+1 < 0 ||
3024 sy >= ssize.height || sy+1 < 0) )
3030 int sx0, sx1, sy0, sy1;
3032 const AT* w = wtab + FXY[dx]*4;
3033 if( borderType == BORDER_REPLICATE )
3035 sx0 = clip(sx, 0, ssize.width);
3036 sx1 = clip(sx+1, 0, ssize.width);
3037 sy0 = clip(sy, 0, ssize.height);
3038 sy1 = clip(sy+1, 0, ssize.height);
3039 v0 = S0[sy0*sstep + sx0];
3040 v1 = S0[sy0*sstep + sx1];
3041 v2 = S0[sy1*sstep + sx0];
3042 v3 = S0[sy1*sstep + sx1];
3046 sx0 = borderInterpolate(sx, ssize.width, borderType);
3047 sx1 = borderInterpolate(sx+1, ssize.width, borderType);
3048 sy0 = borderInterpolate(sy, ssize.height, borderType);
3049 sy1 = borderInterpolate(sy+1, ssize.height, borderType);
3050 v0 = sx0 >= 0 && sy0 >= 0 ? S0[sy0*sstep + sx0] : cval[0];
3051 v1 = sx1 >= 0 && sy0 >= 0 ? S0[sy0*sstep + sx1] : cval[0];
3052 v2 = sx0 >= 0 && sy1 >= 0 ? S0[sy1*sstep + sx0] : cval[0];
3053 v3 = sx1 >= 0 && sy1 >= 0 ? S0[sy1*sstep + sx1] : cval[0];
3055 D[0] = castOp(WT(v0*w[0] + v1*w[1] + v2*w[2] + v3*w[3]));
3059 for( ; dx < X1; dx++, D += cn )
3061 int sx = XY[dx*2], sy = XY[dx*2+1], k;
3062 if( borderType == BORDER_CONSTANT &&
3063 (sx >= ssize.width || sx+1 < 0 ||
3064 sy >= ssize.height || sy+1 < 0) )
3066 for( k = 0; k < cn; k++ )
3071 int sx0, sx1, sy0, sy1;
3072 const T *v0, *v1, *v2, *v3;
3073 const AT* w = wtab + FXY[dx]*4;
3074 if( borderType == BORDER_REPLICATE )
3076 sx0 = clip(sx, 0, ssize.width);
3077 sx1 = clip(sx+1, 0, ssize.width);
3078 sy0 = clip(sy, 0, ssize.height);
3079 sy1 = clip(sy+1, 0, ssize.height);
3080 v0 = S0 + sy0*sstep + sx0*cn;
3081 v1 = S0 + sy0*sstep + sx1*cn;
3082 v2 = S0 + sy1*sstep + sx0*cn;
3083 v3 = S0 + sy1*sstep + sx1*cn;
3085 else if( borderType == BORDER_TRANSPARENT &&
3086 ((unsigned)sx >= (unsigned)(ssize.width-1) ||
3087 (unsigned)sy >= (unsigned)(ssize.height-1)))
3091 sx0 = borderInterpolate(sx, ssize.width, borderType);
3092 sx1 = borderInterpolate(sx+1, ssize.width, borderType);
3093 sy0 = borderInterpolate(sy, ssize.height, borderType);
3094 sy1 = borderInterpolate(sy+1, ssize.height, borderType);
3095 v0 = sx0 >= 0 && sy0 >= 0 ? S0 + sy0*sstep + sx0*cn : &cval[0];
3096 v1 = sx1 >= 0 && sy0 >= 0 ? S0 + sy0*sstep + sx1*cn : &cval[0];
3097 v2 = sx0 >= 0 && sy1 >= 0 ? S0 + sy1*sstep + sx0*cn : &cval[0];
3098 v3 = sx1 >= 0 && sy1 >= 0 ? S0 + sy1*sstep + sx1*cn : &cval[0];
3100 for( k = 0; k < cn; k++ )
3101 D[k] = castOp(WT(v0[k]*w[0] + v1[k]*w[1] + v2[k]*w[2] + v3[k]*w[3]));
3110 template<class CastOp, typename AT, int ONE>
3111 static void remapBicubic( const Mat& _src, Mat& _dst, const Mat& _xy,
3112 const Mat& _fxy, const void* _wtab,
3113 int borderType, const Scalar& _borderValue )
3115 typedef typename CastOp::rtype T;
3116 typedef typename CastOp::type1 WT;
3117 Size ssize = _src.size(), dsize = _dst.size();
3118 int cn = _src.channels();
3119 const AT* wtab = (const AT*)_wtab;
3120 const T* S0 = (const T*)_src.data;
3121 size_t sstep = _src.step/sizeof(S0[0]);
3122 Scalar_<T> cval(saturate_cast<T>(_borderValue[0]),
3123 saturate_cast<T>(_borderValue[1]),
3124 saturate_cast<T>(_borderValue[2]),
3125 saturate_cast<T>(_borderValue[3]));
3128 int borderType1 = borderType != BORDER_TRANSPARENT ? borderType : BORDER_REFLECT_101;
3130 unsigned width1 = std::max(ssize.width-3, 0), height1 = std::max(ssize.height-3, 0);
3132 if( _dst.isContinuous() && _xy.isContinuous() && _fxy.isContinuous() )
3134 dsize.width *= dsize.height;
3138 for( dy = 0; dy < dsize.height; dy++ )
3140 T* D = (T*)(_dst.data + _dst.step*dy);
3141 const short* XY = (const short*)(_xy.data + _xy.step*dy);
3142 const ushort* FXY = (const ushort*)(_fxy.data + _fxy.step*dy);
3144 for( dx = 0; dx < dsize.width; dx++, D += cn )
3146 int sx = XY[dx*2]-1, sy = XY[dx*2+1]-1;
3147 const AT* w = wtab + FXY[dx]*16;
3149 if( (unsigned)sx < width1 && (unsigned)sy < height1 )
3151 const T* S = S0 + sy*sstep + sx*cn;
3152 for( k = 0; k < cn; k++ )
3154 WT sum = S[0]*w[0] + S[cn]*w[1] + S[cn*2]*w[2] + S[cn*3]*w[3];
3156 sum += S[0]*w[4] + S[cn]*w[5] + S[cn*2]*w[6] + S[cn*3]*w[7];
3158 sum += S[0]*w[8] + S[cn]*w[9] + S[cn*2]*w[10] + S[cn*3]*w[11];
3160 sum += S[0]*w[12] + S[cn]*w[13] + S[cn*2]*w[14] + S[cn*3]*w[15];
3168 if( borderType == BORDER_TRANSPARENT &&
3169 ((unsigned)(sx+1) >= (unsigned)ssize.width ||
3170 (unsigned)(sy+1) >= (unsigned)ssize.height) )
3173 if( borderType1 == BORDER_CONSTANT &&
3174 (sx >= ssize.width || sx+4 <= 0 ||
3175 sy >= ssize.height || sy+4 <= 0))
3177 for( k = 0; k < cn; k++ )
3182 for( i = 0; i < 4; i++ )
3184 x[i] = borderInterpolate(sx + i, ssize.width, borderType1)*cn;
3185 y[i] = borderInterpolate(sy + i, ssize.height, borderType1);
3188 for( k = 0; k < cn; k++, S0++, w -= 16 )
3190 WT cv = cval[k], sum = cv*ONE;
3191 for( i = 0; i < 4; i++, w += 4 )
3194 const T* S = S0 + yi*sstep;
3198 sum += (S[x[0]] - cv)*w[0];
3200 sum += (S[x[1]] - cv)*w[1];
3202 sum += (S[x[2]] - cv)*w[2];
3204 sum += (S[x[3]] - cv)*w[3];
3215 template<class CastOp, typename AT, int ONE>
3216 static void remapLanczos4( const Mat& _src, Mat& _dst, const Mat& _xy,
3217 const Mat& _fxy, const void* _wtab,
3218 int borderType, const Scalar& _borderValue )
3220 typedef typename CastOp::rtype T;
3221 typedef typename CastOp::type1 WT;
3222 Size ssize = _src.size(), dsize = _dst.size();
3223 int cn = _src.channels();
3224 const AT* wtab = (const AT*)_wtab;
3225 const T* S0 = (const T*)_src.data;
3226 size_t sstep = _src.step/sizeof(S0[0]);
3227 Scalar_<T> cval(saturate_cast<T>(_borderValue[0]),
3228 saturate_cast<T>(_borderValue[1]),
3229 saturate_cast<T>(_borderValue[2]),
3230 saturate_cast<T>(_borderValue[3]));
3233 int borderType1 = borderType != BORDER_TRANSPARENT ? borderType : BORDER_REFLECT_101;
3235 unsigned width1 = std::max(ssize.width-7, 0), height1 = std::max(ssize.height-7, 0);
3237 if( _dst.isContinuous() && _xy.isContinuous() && _fxy.isContinuous() )
3239 dsize.width *= dsize.height;
3243 for( dy = 0; dy < dsize.height; dy++ )
3245 T* D = (T*)(_dst.data + _dst.step*dy);
3246 const short* XY = (const short*)(_xy.data + _xy.step*dy);
3247 const ushort* FXY = (const ushort*)(_fxy.data + _fxy.step*dy);
3249 for( dx = 0; dx < dsize.width; dx++, D += cn )
3251 int sx = XY[dx*2]-3, sy = XY[dx*2+1]-3;
3252 const AT* w = wtab + FXY[dx]*64;
3253 const T* S = S0 + sy*sstep + sx*cn;
3255 if( (unsigned)sx < width1 && (unsigned)sy < height1 )
3257 for( k = 0; k < cn; k++ )
3260 for( int r = 0; r < 8; r++, S += sstep, w += 8 )
3261 sum += S[0]*w[0] + S[cn]*w[1] + S[cn*2]*w[2] + S[cn*3]*w[3] +
3262 S[cn*4]*w[4] + S[cn*5]*w[5] + S[cn*6]*w[6] + S[cn*7]*w[7];
3271 if( borderType == BORDER_TRANSPARENT &&
3272 ((unsigned)(sx+3) >= (unsigned)ssize.width ||
3273 (unsigned)(sy+3) >= (unsigned)ssize.height) )
3276 if( borderType1 == BORDER_CONSTANT &&
3277 (sx >= ssize.width || sx+8 <= 0 ||
3278 sy >= ssize.height || sy+8 <= 0))
3280 for( k = 0; k < cn; k++ )
3285 for( i = 0; i < 8; i++ )
3287 x[i] = borderInterpolate(sx + i, ssize.width, borderType1)*cn;
3288 y[i] = borderInterpolate(sy + i, ssize.height, borderType1);
3291 for( k = 0; k < cn; k++, S0++, w -= 64 )
3293 WT cv = cval[k], sum = cv*ONE;
3294 for( i = 0; i < 8; i++, w += 8 )
3297 const T* S1 = S0 + yi*sstep;
3301 sum += (S1[x[0]] - cv)*w[0];
3303 sum += (S1[x[1]] - cv)*w[1];
3305 sum += (S1[x[2]] - cv)*w[2];
3307 sum += (S1[x[3]] - cv)*w[3];
3309 sum += (S1[x[4]] - cv)*w[4];
3311 sum += (S1[x[5]] - cv)*w[5];
3313 sum += (S1[x[6]] - cv)*w[6];
3315 sum += (S1[x[7]] - cv)*w[7];
3326 typedef void (*RemapNNFunc)(const Mat& _src, Mat& _dst, const Mat& _xy,
3327 int borderType, const Scalar& _borderValue );
3329 typedef void (*RemapFunc)(const Mat& _src, Mat& _dst, const Mat& _xy,
3330 const Mat& _fxy, const void* _wtab,
3331 int borderType, const Scalar& _borderValue);
3333 class RemapInvoker :
3334 public ParallelLoopBody
3337 RemapInvoker(const Mat& _src, Mat& _dst, const Mat *_m1,
3338 const Mat *_m2, int _borderType, const Scalar &_borderValue,
3339 int _planar_input, RemapNNFunc _nnfunc, RemapFunc _ifunc, const void *_ctab) :
3340 ParallelLoopBody(), src(&_src), dst(&_dst), m1(_m1), m2(_m2),
3341 borderType(_borderType), borderValue(_borderValue),
3342 planar_input(_planar_input), nnfunc(_nnfunc), ifunc(_ifunc), ctab(_ctab)
3346 virtual void operator() (const Range& range) const
3349 const int buf_size = 1 << 14;
3350 int brows0 = std::min(128, dst->rows), map_depth = m1->depth();
3351 int bcols0 = std::min(buf_size/brows0, dst->cols);
3352 brows0 = std::min(buf_size/bcols0, dst->rows);
3354 bool useSIMD = checkHardwareSupport(CV_CPU_SSE2);
3357 Mat _bufxy(brows0, bcols0, CV_16SC2), _bufa;
3359 _bufa.create(brows0, bcols0, CV_16UC1);
3361 for( y = range.start; y < range.end; y += brows0 )
3363 for( x = 0; x < dst->cols; x += bcols0 )
3365 int brows = std::min(brows0, range.end - y);
3366 int bcols = std::min(bcols0, dst->cols - x);
3367 Mat dpart(*dst, Rect(x, y, bcols, brows));
3368 Mat bufxy(_bufxy, Rect(0, 0, bcols, brows));
3372 if( m1->type() == CV_16SC2 && !m2->data ) // the data is already in the right format
3373 bufxy = (*m1)(Rect(x, y, bcols, brows));
3374 else if( map_depth != CV_32F )
3376 for( y1 = 0; y1 < brows; y1++ )
3378 short* XY = (short*)(bufxy.data + bufxy.step*y1);
3379 const short* sXY = (const short*)(m1->data + m1->step*(y+y1)) + x*2;
3380 const ushort* sA = (const ushort*)(m2->data + m2->step*(y+y1)) + x;
3382 for( x1 = 0; x1 < bcols; x1++ )
3384 int a = sA[x1] & (INTER_TAB_SIZE2-1);
3385 XY[x1*2] = sXY[x1*2] + NNDeltaTab_i[a][0];
3386 XY[x1*2+1] = sXY[x1*2+1] + NNDeltaTab_i[a][1];
3390 else if( !planar_input )
3391 (*m1)(Rect(x, y, bcols, brows)).convertTo(bufxy, bufxy.depth());
3394 for( y1 = 0; y1 < brows; y1++ )
3396 short* XY = (short*)(bufxy.data + bufxy.step*y1);
3397 const float* sX = (const float*)(m1->data + m1->step*(y+y1)) + x;
3398 const float* sY = (const float*)(m2->data + m2->step*(y+y1)) + x;
3404 for( ; x1 <= bcols - 8; x1 += 8 )
3406 __m128 fx0 = _mm_loadu_ps(sX + x1);
3407 __m128 fx1 = _mm_loadu_ps(sX + x1 + 4);
3408 __m128 fy0 = _mm_loadu_ps(sY + x1);
3409 __m128 fy1 = _mm_loadu_ps(sY + x1 + 4);
3410 __m128i ix0 = _mm_cvtps_epi32(fx0);
3411 __m128i ix1 = _mm_cvtps_epi32(fx1);
3412 __m128i iy0 = _mm_cvtps_epi32(fy0);
3413 __m128i iy1 = _mm_cvtps_epi32(fy1);
3414 ix0 = _mm_packs_epi32(ix0, ix1);
3415 iy0 = _mm_packs_epi32(iy0, iy1);
3416 ix1 = _mm_unpacklo_epi16(ix0, iy0);
3417 iy1 = _mm_unpackhi_epi16(ix0, iy0);
3418 _mm_storeu_si128((__m128i*)(XY + x1*2), ix1);
3419 _mm_storeu_si128((__m128i*)(XY + x1*2 + 8), iy1);
3424 for( ; x1 < bcols; x1++ )
3426 XY[x1*2] = saturate_cast<short>(sX[x1]);
3427 XY[x1*2+1] = saturate_cast<short>(sY[x1]);
3431 nnfunc( *src, dpart, bufxy, borderType, borderValue );
3435 Mat bufa(_bufa, Rect(0, 0, bcols, brows));
3436 for( y1 = 0; y1 < brows; y1++ )
3438 short* XY = (short*)(bufxy.data + bufxy.step*y1);
3439 ushort* A = (ushort*)(bufa.data + bufa.step*y1);
3441 if( m1->type() == CV_16SC2 && (m2->type() == CV_16UC1 || m2->type() == CV_16SC1) )
3443 bufxy = (*m1)(Rect(x, y, bcols, brows));
3445 const ushort* sA = (const ushort*)(m2->data + m2->step*(y+y1)) + x;
3446 for( x1 = 0; x1 < bcols; x1++ )
3447 A[x1] = (ushort)(sA[x1] & (INTER_TAB_SIZE2-1));
3449 else if( planar_input )
3451 const float* sX = (const float*)(m1->data + m1->step*(y+y1)) + x;
3452 const float* sY = (const float*)(m2->data + m2->step*(y+y1)) + x;
3458 __m128 scale = _mm_set1_ps((float)INTER_TAB_SIZE);
3459 __m128i mask = _mm_set1_epi32(INTER_TAB_SIZE-1);
3460 for( ; x1 <= bcols - 8; x1 += 8 )
3462 __m128 fx0 = _mm_loadu_ps(sX + x1);
3463 __m128 fx1 = _mm_loadu_ps(sX + x1 + 4);
3464 __m128 fy0 = _mm_loadu_ps(sY + x1);
3465 __m128 fy1 = _mm_loadu_ps(sY + x1 + 4);
3466 __m128i ix0 = _mm_cvtps_epi32(_mm_mul_ps(fx0, scale));
3467 __m128i ix1 = _mm_cvtps_epi32(_mm_mul_ps(fx1, scale));
3468 __m128i iy0 = _mm_cvtps_epi32(_mm_mul_ps(fy0, scale));
3469 __m128i iy1 = _mm_cvtps_epi32(_mm_mul_ps(fy1, scale));
3470 __m128i mx0 = _mm_and_si128(ix0, mask);
3471 __m128i mx1 = _mm_and_si128(ix1, mask);
3472 __m128i my0 = _mm_and_si128(iy0, mask);
3473 __m128i my1 = _mm_and_si128(iy1, mask);
3474 mx0 = _mm_packs_epi32(mx0, mx1);
3475 my0 = _mm_packs_epi32(my0, my1);
3476 my0 = _mm_slli_epi16(my0, INTER_BITS);
3477 mx0 = _mm_or_si128(mx0, my0);
3478 _mm_storeu_si128((__m128i*)(A + x1), mx0);
3479 ix0 = _mm_srai_epi32(ix0, INTER_BITS);
3480 ix1 = _mm_srai_epi32(ix1, INTER_BITS);
3481 iy0 = _mm_srai_epi32(iy0, INTER_BITS);
3482 iy1 = _mm_srai_epi32(iy1, INTER_BITS);
3483 ix0 = _mm_packs_epi32(ix0, ix1);
3484 iy0 = _mm_packs_epi32(iy0, iy1);
3485 ix1 = _mm_unpacklo_epi16(ix0, iy0);
3486 iy1 = _mm_unpackhi_epi16(ix0, iy0);
3487 _mm_storeu_si128((__m128i*)(XY + x1*2), ix1);
3488 _mm_storeu_si128((__m128i*)(XY + x1*2 + 8), iy1);
3493 for( ; x1 < bcols; x1++ )
3495 int sx = cvRound(sX[x1]*INTER_TAB_SIZE);
3496 int sy = cvRound(sY[x1]*INTER_TAB_SIZE);
3497 int v = (sy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (sx & (INTER_TAB_SIZE-1));
3498 XY[x1*2] = saturate_cast<short>(sx >> INTER_BITS);
3499 XY[x1*2+1] = saturate_cast<short>(sy >> INTER_BITS);
3505 const float* sXY = (const float*)(m1->data + m1->step*(y+y1)) + x*2;
3507 for( x1 = 0; x1 < bcols; x1++ )
3509 int sx = cvRound(sXY[x1*2]*INTER_TAB_SIZE);
3510 int sy = cvRound(sXY[x1*2+1]*INTER_TAB_SIZE);
3511 int v = (sy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (sx & (INTER_TAB_SIZE-1));
3512 XY[x1*2] = saturate_cast<short>(sx >> INTER_BITS);
3513 XY[x1*2+1] = saturate_cast<short>(sy >> INTER_BITS);
3518 ifunc(*src, dpart, bufxy, bufa, ctab, borderType, borderValue);
3537 static bool ocl_remap(InputArray _src, OutputArray _dst, InputArray _map1, InputArray _map2,
3538 int interpolation, int borderType, const Scalar& borderValue)
3540 int cn = _src.channels(), type = _src.type(), depth = _src.depth();
3542 if (borderType == BORDER_TRANSPARENT || cn == 3 || !(interpolation == INTER_LINEAR || interpolation == INTER_NEAREST)
3543 || _map1.type() == CV_16SC1 || _map2.type() == CV_16SC1)
3546 UMat src = _src.getUMat(), map1 = _map1.getUMat(), map2 = _map2.getUMat();
3548 if( (map1.type() == CV_16SC2 && (map2.type() == CV_16UC1 || map2.empty())) ||
3549 (map2.type() == CV_16SC2 && (map1.type() == CV_16UC1 || map1.empty())) )
3551 if (map1.type() != CV_16SC2)
3552 std::swap(map1, map2);
3555 CV_Assert( map1.type() == CV_32FC2 || (map1.type() == CV_32FC1 && map2.type() == CV_32FC1) );
3557 _dst.create(map1.size(), type);
3558 UMat dst = _dst.getUMat();
3560 String kernelName = "remap";
3561 if (map1.type() == CV_32FC2 && map2.empty())
3562 kernelName += "_32FC2";
3563 else if (map1.type() == CV_16SC2)
3565 kernelName += "_16SC2";
3567 kernelName += "_16UC1";
3569 else if (map1.type() == CV_32FC1 && map2.type() == CV_32FC1)
3570 kernelName += "_2_32FC1";
3572 CV_Error(Error::StsBadArg, "Unsupported map types");
3574 static const char * const interMap[] = { "INTER_NEAREST", "INTER_LINEAR", "INTER_CUBIC", "INTER_LINEAR", "INTER_LANCZOS" };
3575 static const char * const borderMap[] = { "BORDER_CONSTANT", "BORDER_REPLICATE", "BORDER_REFLECT", "BORDER_WRAP",
3576 "BORDER_REFLECT_101", "BORDER_TRANSPARENT" };
3577 String buildOptions = format("-D %s -D %s -D T=%s", interMap[interpolation], borderMap[borderType], ocl::typeToStr(type));
3579 if (interpolation != INTER_NEAREST)
3582 int wdepth = std::max(CV_32F, dst.depth());
3583 buildOptions = buildOptions
3584 + format(" -D WT=%s -D convertToT=%s -D convertToWT=%s"
3585 " -D convertToWT2=%s -D WT2=%s",
3586 ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)),
3587 ocl::convertTypeStr(wdepth, depth, cn, cvt[0]),
3588 ocl::convertTypeStr(depth, wdepth, cn, cvt[1]),
3589 ocl::convertTypeStr(CV_32S, wdepth, 2, cvt[2]),
3590 ocl::typeToStr(CV_MAKE_TYPE(wdepth, 2)));
3593 ocl::Kernel k(kernelName.c_str(), ocl::imgproc::remap_oclsrc, buildOptions);
3595 Mat scalar(1, 1, type, borderValue);
3596 ocl::KernelArg srcarg = ocl::KernelArg::ReadOnly(src), dstarg = ocl::KernelArg::WriteOnly(dst),
3597 map1arg = ocl::KernelArg::ReadOnlyNoSize(map1),
3598 scalararg = ocl::KernelArg::Constant((void*)scalar.data, scalar.elemSize());
3601 k.args(srcarg, dstarg, map1arg, scalararg);
3603 k.args(srcarg, dstarg, map1arg, ocl::KernelArg::ReadOnlyNoSize(map2), scalararg);
3605 size_t globalThreads[2] = { dst.cols, dst.rows };
3606 return k.run(2, globalThreads, NULL, false);
3613 void cv::remap( InputArray _src, OutputArray _dst,
3614 InputArray _map1, InputArray _map2,
3615 int interpolation, int borderType, const Scalar& borderValue )
3617 static RemapNNFunc nn_tab[] =
3619 remapNearest<uchar>, remapNearest<schar>, remapNearest<ushort>, remapNearest<short>,
3620 remapNearest<int>, remapNearest<float>, remapNearest<double>, 0
3623 static RemapFunc linear_tab[] =
3625 remapBilinear<FixedPtCast<int, uchar, INTER_REMAP_COEF_BITS>, RemapVec_8u, short>, 0,
3626 remapBilinear<Cast<float, ushort>, RemapNoVec, float>,
3627 remapBilinear<Cast<float, short>, RemapNoVec, float>, 0,
3628 remapBilinear<Cast<float, float>, RemapNoVec, float>,
3629 remapBilinear<Cast<double, double>, RemapNoVec, float>, 0
3632 static RemapFunc cubic_tab[] =
3634 remapBicubic<FixedPtCast<int, uchar, INTER_REMAP_COEF_BITS>, short, INTER_REMAP_COEF_SCALE>, 0,
3635 remapBicubic<Cast<float, ushort>, float, 1>,
3636 remapBicubic<Cast<float, short>, float, 1>, 0,
3637 remapBicubic<Cast<float, float>, float, 1>,
3638 remapBicubic<Cast<double, double>, float, 1>, 0
3641 static RemapFunc lanczos4_tab[] =
3643 remapLanczos4<FixedPtCast<int, uchar, INTER_REMAP_COEF_BITS>, short, INTER_REMAP_COEF_SCALE>, 0,
3644 remapLanczos4<Cast<float, ushort>, float, 1>,
3645 remapLanczos4<Cast<float, short>, float, 1>, 0,
3646 remapLanczos4<Cast<float, float>, float, 1>,
3647 remapLanczos4<Cast<double, double>, float, 1>, 0
3650 CV_Assert( _map1.size().area() > 0 );
3651 CV_Assert( _map2.empty() || (_map2.size() == _map1.size()));
3653 CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
3654 ocl_remap(_src, _dst, _map1, _map2, interpolation, borderType, borderValue))
3656 Mat src = _src.getMat(), map1 = _map1.getMat(), map2 = _map2.getMat();
3657 _dst.create( map1.size(), src.type() );
3658 Mat dst = _dst.getMat();
3659 if( dst.data == src.data )
3662 int depth = src.depth();
3663 RemapNNFunc nnfunc = 0;
3664 RemapFunc ifunc = 0;
3665 const void* ctab = 0;
3666 bool fixpt = depth == CV_8U;
3667 bool planar_input = false;
3669 if( interpolation == INTER_NEAREST )
3671 nnfunc = nn_tab[depth];
3672 CV_Assert( nnfunc != 0 );
3676 if( interpolation == INTER_AREA )
3677 interpolation = INTER_LINEAR;
3679 if( interpolation == INTER_LINEAR )
3680 ifunc = linear_tab[depth];
3681 else if( interpolation == INTER_CUBIC )
3682 ifunc = cubic_tab[depth];
3683 else if( interpolation == INTER_LANCZOS4 )
3684 ifunc = lanczos4_tab[depth];
3686 CV_Error( CV_StsBadArg, "Unknown interpolation method" );
3687 CV_Assert( ifunc != 0 );
3688 ctab = initInterTab2D( interpolation, fixpt );
3691 const Mat *m1 = &map1, *m2 = &map2;
3693 if( (map1.type() == CV_16SC2 && (map2.type() == CV_16UC1 || map2.type() == CV_16SC1 || !map2.data)) ||
3694 (map2.type() == CV_16SC2 && (map1.type() == CV_16UC1 || map1.type() == CV_16SC1 || !map1.data)) )
3696 if( map1.type() != CV_16SC2 )
3701 CV_Assert( ((map1.type() == CV_32FC2 || map1.type() == CV_16SC2) && !map2.data) ||
3702 (map1.type() == CV_32FC1 && map2.type() == CV_32FC1) );
3703 planar_input = map1.channels() == 1;
3706 RemapInvoker invoker(src, dst, m1, m2,
3707 borderType, borderValue, planar_input, nnfunc, ifunc,
3709 parallel_for_(Range(0, dst.rows), invoker, dst.total()/(double)(1<<16));
3713 void cv::convertMaps( InputArray _map1, InputArray _map2,
3714 OutputArray _dstmap1, OutputArray _dstmap2,
3715 int dstm1type, bool nninterpolate )
3717 Mat map1 = _map1.getMat(), map2 = _map2.getMat(), dstmap1, dstmap2;
3718 Size size = map1.size();
3719 const Mat *m1 = &map1, *m2 = &map2;
3720 int m1type = m1->type(), m2type = m2->type();
3722 CV_Assert( (m1type == CV_16SC2 && (nninterpolate || m2type == CV_16UC1 || m2type == CV_16SC1)) ||
3723 (m2type == CV_16SC2 && (nninterpolate || m1type == CV_16UC1 || m1type == CV_16SC1)) ||
3724 (m1type == CV_32FC1 && m2type == CV_32FC1) ||
3725 (m1type == CV_32FC2 && !m2->data) );
3727 if( m2type == CV_16SC2 )
3729 std::swap( m1, m2 );
3730 std::swap( m1type, m2type );
3733 if( dstm1type <= 0 )
3734 dstm1type = m1type == CV_16SC2 ? CV_32FC2 : CV_16SC2;
3735 CV_Assert( dstm1type == CV_16SC2 || dstm1type == CV_32FC1 || dstm1type == CV_32FC2 );
3736 _dstmap1.create( size, dstm1type );
3737 dstmap1 = _dstmap1.getMat();
3739 if( !nninterpolate && dstm1type != CV_32FC2 )
3741 _dstmap2.create( size, dstm1type == CV_16SC2 ? CV_16UC1 : CV_32FC1 );
3742 dstmap2 = _dstmap2.getMat();
3747 if( m1type == dstm1type || (nninterpolate &&
3748 ((m1type == CV_16SC2 && dstm1type == CV_32FC2) ||
3749 (m1type == CV_32FC2 && dstm1type == CV_16SC2))) )
3751 m1->convertTo( dstmap1, dstmap1.type() );
3752 if( dstmap2.data && dstmap2.type() == m2->type() )
3753 m2->copyTo( dstmap2 );
3757 if( m1type == CV_32FC1 && dstm1type == CV_32FC2 )
3759 Mat vdata[] = { *m1, *m2 };
3760 merge( vdata, 2, dstmap1 );
3764 if( m1type == CV_32FC2 && dstm1type == CV_32FC1 )
3766 Mat mv[] = { dstmap1, dstmap2 };
3771 if( m1->isContinuous() && (!m2->data || m2->isContinuous()) &&
3772 dstmap1.isContinuous() && (!dstmap2.data || dstmap2.isContinuous()) )
3774 size.width *= size.height;
3778 const float scale = 1.f/INTER_TAB_SIZE;
3780 for( y = 0; y < size.height; y++ )
3782 const float* src1f = (const float*)(m1->data + m1->step*y);
3783 const float* src2f = (const float*)(m2->data + m2->step*y);
3784 const short* src1 = (const short*)src1f;
3785 const ushort* src2 = (const ushort*)src2f;
3787 float* dst1f = (float*)(dstmap1.data + dstmap1.step*y);
3788 float* dst2f = (float*)(dstmap2.data + dstmap2.step*y);
3789 short* dst1 = (short*)dst1f;
3790 ushort* dst2 = (ushort*)dst2f;
3792 if( m1type == CV_32FC1 && dstm1type == CV_16SC2 )
3795 for( x = 0; x < size.width; x++ )
3797 dst1[x*2] = saturate_cast<short>(src1f[x]);
3798 dst1[x*2+1] = saturate_cast<short>(src2f[x]);
3801 for( x = 0; x < size.width; x++ )
3803 int ix = saturate_cast<int>(src1f[x]*INTER_TAB_SIZE);
3804 int iy = saturate_cast<int>(src2f[x]*INTER_TAB_SIZE);
3805 dst1[x*2] = saturate_cast<short>(ix >> INTER_BITS);
3806 dst1[x*2+1] = saturate_cast<short>(iy >> INTER_BITS);
3807 dst2[x] = (ushort)((iy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (ix & (INTER_TAB_SIZE-1)));
3810 else if( m1type == CV_32FC2 && dstm1type == CV_16SC2 )
3813 for( x = 0; x < size.width; x++ )
3815 dst1[x*2] = saturate_cast<short>(src1f[x*2]);
3816 dst1[x*2+1] = saturate_cast<short>(src1f[x*2+1]);
3819 for( x = 0; x < size.width; x++ )
3821 int ix = saturate_cast<int>(src1f[x*2]*INTER_TAB_SIZE);
3822 int iy = saturate_cast<int>(src1f[x*2+1]*INTER_TAB_SIZE);
3823 dst1[x*2] = saturate_cast<short>(ix >> INTER_BITS);
3824 dst1[x*2+1] = saturate_cast<short>(iy >> INTER_BITS);
3825 dst2[x] = (ushort)((iy & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE + (ix & (INTER_TAB_SIZE-1)));
3828 else if( m1type == CV_16SC2 && dstm1type == CV_32FC1 )
3830 for( x = 0; x < size.width; x++ )
3832 int fxy = src2 ? src2[x] & (INTER_TAB_SIZE2-1) : 0;
3833 dst1f[x] = src1[x*2] + (fxy & (INTER_TAB_SIZE-1))*scale;
3834 dst2f[x] = src1[x*2+1] + (fxy >> INTER_BITS)*scale;
3837 else if( m1type == CV_16SC2 && dstm1type == CV_32FC2 )
3839 for( x = 0; x < size.width; x++ )
3841 int fxy = src2 ? src2[x] & (INTER_TAB_SIZE2-1): 0;
3842 dst1f[x*2] = src1[x*2] + (fxy & (INTER_TAB_SIZE-1))*scale;
3843 dst1f[x*2+1] = src1[x*2+1] + (fxy >> INTER_BITS)*scale;
3847 CV_Error( CV_StsNotImplemented, "Unsupported combination of input/output matrices" );
3855 class warpAffineInvoker :
3856 public ParallelLoopBody
3859 warpAffineInvoker(const Mat &_src, Mat &_dst, int _interpolation, int _borderType,
3860 const Scalar &_borderValue, int *_adelta, int *_bdelta, double *_M) :
3861 ParallelLoopBody(), src(_src), dst(_dst), interpolation(_interpolation),
3862 borderType(_borderType), borderValue(_borderValue), adelta(_adelta), bdelta(_bdelta),
3867 virtual void operator() (const Range& range) const
3869 const int BLOCK_SZ = 64;
3870 short XY[BLOCK_SZ*BLOCK_SZ*2], A[BLOCK_SZ*BLOCK_SZ];
3871 const int AB_BITS = MAX(10, (int)INTER_BITS);
3872 const int AB_SCALE = 1 << AB_BITS;
3873 int round_delta = interpolation == INTER_NEAREST ? AB_SCALE/2 : AB_SCALE/INTER_TAB_SIZE/2, x, y, x1, y1;
3875 bool useSIMD = checkHardwareSupport(CV_CPU_SSE2);
3878 int bh0 = std::min(BLOCK_SZ/2, dst.rows);
3879 int bw0 = std::min(BLOCK_SZ*BLOCK_SZ/bh0, dst.cols);
3880 bh0 = std::min(BLOCK_SZ*BLOCK_SZ/bw0, dst.rows);
3882 for( y = range.start; y < range.end; y += bh0 )
3884 for( x = 0; x < dst.cols; x += bw0 )
3886 int bw = std::min( bw0, dst.cols - x);
3887 int bh = std::min( bh0, range.end - y);
3889 Mat _XY(bh, bw, CV_16SC2, XY), matA;
3890 Mat dpart(dst, Rect(x, y, bw, bh));
3892 for( y1 = 0; y1 < bh; y1++ )
3894 short* xy = XY + y1*bw*2;
3895 int X0 = saturate_cast<int>((M[1]*(y + y1) + M[2])*AB_SCALE) + round_delta;
3896 int Y0 = saturate_cast<int>((M[4]*(y + y1) + M[5])*AB_SCALE) + round_delta;
3898 if( interpolation == INTER_NEAREST )
3899 for( x1 = 0; x1 < bw; x1++ )
3901 int X = (X0 + adelta[x+x1]) >> AB_BITS;
3902 int Y = (Y0 + bdelta[x+x1]) >> AB_BITS;
3903 xy[x1*2] = saturate_cast<short>(X);
3904 xy[x1*2+1] = saturate_cast<short>(Y);
3908 short* alpha = A + y1*bw;
3913 __m128i fxy_mask = _mm_set1_epi32(INTER_TAB_SIZE - 1);
3914 __m128i XX = _mm_set1_epi32(X0), YY = _mm_set1_epi32(Y0);
3915 for( ; x1 <= bw - 8; x1 += 8 )
3917 __m128i tx0, tx1, ty0, ty1;
3918 tx0 = _mm_add_epi32(_mm_loadu_si128((const __m128i*)(adelta + x + x1)), XX);
3919 ty0 = _mm_add_epi32(_mm_loadu_si128((const __m128i*)(bdelta + x + x1)), YY);
3920 tx1 = _mm_add_epi32(_mm_loadu_si128((const __m128i*)(adelta + x + x1 + 4)), XX);
3921 ty1 = _mm_add_epi32(_mm_loadu_si128((const __m128i*)(bdelta + x + x1 + 4)), YY);
3923 tx0 = _mm_srai_epi32(tx0, AB_BITS - INTER_BITS);
3924 ty0 = _mm_srai_epi32(ty0, AB_BITS - INTER_BITS);
3925 tx1 = _mm_srai_epi32(tx1, AB_BITS - INTER_BITS);
3926 ty1 = _mm_srai_epi32(ty1, AB_BITS - INTER_BITS);
3928 __m128i fx_ = _mm_packs_epi32(_mm_and_si128(tx0, fxy_mask),
3929 _mm_and_si128(tx1, fxy_mask));
3930 __m128i fy_ = _mm_packs_epi32(_mm_and_si128(ty0, fxy_mask),
3931 _mm_and_si128(ty1, fxy_mask));
3932 tx0 = _mm_packs_epi32(_mm_srai_epi32(tx0, INTER_BITS),
3933 _mm_srai_epi32(tx1, INTER_BITS));
3934 ty0 = _mm_packs_epi32(_mm_srai_epi32(ty0, INTER_BITS),
3935 _mm_srai_epi32(ty1, INTER_BITS));
3936 fx_ = _mm_adds_epi16(fx_, _mm_slli_epi16(fy_, INTER_BITS));
3938 _mm_storeu_si128((__m128i*)(xy + x1*2), _mm_unpacklo_epi16(tx0, ty0));
3939 _mm_storeu_si128((__m128i*)(xy + x1*2 + 8), _mm_unpackhi_epi16(tx0, ty0));
3940 _mm_storeu_si128((__m128i*)(alpha + x1), fx_);
3944 for( ; x1 < bw; x1++ )
3946 int X = (X0 + adelta[x+x1]) >> (AB_BITS - INTER_BITS);
3947 int Y = (Y0 + bdelta[x+x1]) >> (AB_BITS - INTER_BITS);
3948 xy[x1*2] = saturate_cast<short>(X >> INTER_BITS);
3949 xy[x1*2+1] = saturate_cast<short>(Y >> INTER_BITS);
3950 alpha[x1] = (short)((Y & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE +
3951 (X & (INTER_TAB_SIZE-1)));
3956 if( interpolation == INTER_NEAREST )
3957 remap( src, dpart, _XY, Mat(), interpolation, borderType, borderValue );
3960 Mat _matA(bh, bw, CV_16U, A);
3961 remap( src, dpart, _XY, _matA, interpolation, borderType, borderValue );
3970 int interpolation, borderType;
3972 int *adelta, *bdelta;
3976 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
3977 class IPPwarpAffineInvoker :
3978 public ParallelLoopBody
3981 IPPwarpAffineInvoker(Mat &_src, Mat &_dst, double (&_coeffs)[2][3], int &_interpolation, int &_borderType, const Scalar &_borderValue, ippiWarpAffineBackFunc _func, bool *_ok) :
3982 ParallelLoopBody(), src(_src), dst(_dst), mode(_interpolation), coeffs(_coeffs), borderType(_borderType), borderValue(_borderValue), func(_func), ok(_ok)
3987 virtual void operator() (const Range& range) const
3989 IppiSize srcsize = { src.cols, src.rows };
3990 IppiRect srcroi = { 0, 0, src.cols, src.rows };
3991 IppiRect dstroi = { 0, range.start, dst.cols, range.end - range.start };
3992 int cnn = src.channels();
3993 if( borderType == BORDER_CONSTANT )
3995 IppiSize setSize = { dst.cols, range.end - range.start };
3996 void *dataPointer = dst.data + dst.step[0] * range.start;
3997 if( !IPPSet( borderValue, dataPointer, (int)dst.step[0], setSize, cnn, src.depth() ) )
4003 if( func( src.data, srcsize, (int)src.step[0], srcroi, dst.data, (int)dst.step[0], dstroi, coeffs, mode ) < 0) ////Aug 2013: problem in IPP 7.1, 8.0 : sometimes function return ippStsCoeffErr
4009 double (&coeffs)[2][3];
4013 ippiWarpAffineBackFunc func;
4015 const IPPwarpAffineInvoker& operator= (const IPPwarpAffineInvoker&);
4021 enum { OCL_OP_PERSPECTIVE = 1, OCL_OP_AFFINE = 0 };
4023 static bool ocl_warpTransform(InputArray _src, OutputArray _dst, InputArray _M0,
4024 Size dsize, int flags, int borderType, const Scalar& borderValue,
4027 CV_Assert(op_type == OCL_OP_AFFINE || op_type == OCL_OP_PERSPECTIVE);
4029 int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
4030 double doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
4032 int interpolation = flags & INTER_MAX;
4033 if( interpolation == INTER_AREA )
4034 interpolation = INTER_LINEAR;
4036 if ( !(borderType == cv::BORDER_CONSTANT &&
4037 (interpolation == cv::INTER_NEAREST || interpolation == cv::INTER_LINEAR || interpolation == cv::INTER_CUBIC)) ||
4038 (!doubleSupport && depth == CV_64F) || cn > 4)
4041 const char * const interpolationMap[3] = { "NEAREST", "LINEAR", "CUBIC" };
4042 ocl::ProgramSource program = op_type == OCL_OP_AFFINE ?
4043 ocl::imgproc::warp_affine_oclsrc : ocl::imgproc::warp_perspective_oclsrc;
4044 const char * const kernelName = op_type == OCL_OP_AFFINE ? "warpAffine" : "warpPerspective";
4046 int scalarcn = cn == 3 ? 4 : cn;
4047 int wdepth = interpolation == INTER_NEAREST ? depth : std::max(CV_32S, depth);
4048 int sctype = CV_MAKETYPE(wdepth, scalarcn);
4052 if (interpolation == INTER_NEAREST)
4054 opts = format("-D INTER_NEAREST -D T=%s%s -D T1=%s -D ST=%s -D cn=%d", ocl::typeToStr(type),
4055 doubleSupport ? " -D DOUBLE_SUPPORT" : "",
4056 ocl::typeToStr(CV_MAT_DEPTH(type)),
4057 ocl::typeToStr(sctype),
4063 opts = format("-D INTER_%s -D T=%s -D T1=%s -D ST=%s -D WT=%s -D depth=%d -D convertToWT=%s -D convertToT=%s%s -D cn=%d",
4064 interpolationMap[interpolation], ocl::typeToStr(type),
4065 ocl::typeToStr(CV_MAT_DEPTH(type)),
4066 ocl::typeToStr(sctype),
4067 ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)), depth,
4068 ocl::convertTypeStr(depth, wdepth, cn, cvt[0]),
4069 ocl::convertTypeStr(wdepth, depth, cn, cvt[1]),
4070 doubleSupport ? " -D DOUBLE_SUPPORT" : "", cn);
4073 k.create(kernelName, program, opts);
4077 double borderBuf[] = {0, 0, 0, 0};
4078 scalarToRawData(borderValue, borderBuf, sctype);
4080 UMat src = _src.getUMat(), M0;
4081 _dst.create( dsize.area() == 0 ? src.size() : dsize, src.type() );
4082 UMat dst = _dst.getUMat();
4085 int matRows = (op_type == OCL_OP_AFFINE ? 2 : 3);
4086 Mat matM(matRows, 3, CV_64F, M), M1 = _M0.getMat();
4087 CV_Assert( (M1.type() == CV_32F || M1.type() == CV_64F) &&
4088 M1.rows == matRows && M1.cols == 3 );
4089 M1.convertTo(matM, matM.type());
4091 if( !(flags & WARP_INVERSE_MAP) )
4093 if (op_type == OCL_OP_PERSPECTIVE)
4097 double D = M[0]*M[4] - M[1]*M[3];
4098 D = D != 0 ? 1./D : 0;
4099 double A11 = M[4]*D, A22=M[0]*D;
4100 M[0] = A11; M[1] *= -D;
4101 M[3] *= -D; M[4] = A22;
4102 double b1 = -M[0]*M[2] - M[1]*M[5];
4103 double b2 = -M[3]*M[2] - M[4]*M[5];
4104 M[2] = b1; M[5] = b2;
4107 matM.convertTo(M0, doubleSupport ? CV_64F : CV_32F);
4109 k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst), ocl::KernelArg::PtrReadOnly(M0),
4110 ocl::KernelArg(0, 0, 0, borderBuf, CV_ELEM_SIZE(sctype)));
4112 size_t globalThreads[2] = { dst.cols, dst.rows };
4113 return k.run(2, globalThreads, NULL, false);
4121 void cv::warpAffine( InputArray _src, OutputArray _dst,
4122 InputArray _M0, Size dsize,
4123 int flags, int borderType, const Scalar& borderValue )
4125 CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
4126 ocl_warpTransform(_src, _dst, _M0, dsize, flags, borderType,
4127 borderValue, OCL_OP_AFFINE))
4129 Mat src = _src.getMat(), M0 = _M0.getMat();
4130 _dst.create( dsize.area() == 0 ? src.size() : dsize, src.type() );
4131 Mat dst = _dst.getMat();
4132 CV_Assert( src.cols > 0 && src.rows > 0 );
4133 if( dst.data == src.data )
4137 Mat matM(2, 3, CV_64F, M);
4138 int interpolation = flags & INTER_MAX;
4139 if( interpolation == INTER_AREA )
4140 interpolation = INTER_LINEAR;
4142 CV_Assert( (M0.type() == CV_32F || M0.type() == CV_64F) && M0.rows == 2 && M0.cols == 3 );
4143 M0.convertTo(matM, matM.type());
4145 #ifdef HAVE_TEGRA_OPTIMIZATION
4146 if( tegra::warpAffine(src, dst, M, flags, borderType, borderValue) )
4150 if( !(flags & WARP_INVERSE_MAP) )
4152 double D = M[0]*M[4] - M[1]*M[3];
4153 D = D != 0 ? 1./D : 0;
4154 double A11 = M[4]*D, A22=M[0]*D;
4155 M[0] = A11; M[1] *= -D;
4156 M[3] *= -D; M[4] = A22;
4157 double b1 = -M[0]*M[2] - M[1]*M[5];
4158 double b2 = -M[3]*M[2] - M[4]*M[5];
4159 M[2] = b1; M[5] = b2;
4163 AutoBuffer<int> _abdelta(dst.cols*2);
4164 int* adelta = &_abdelta[0], *bdelta = adelta + dst.cols;
4165 const int AB_BITS = MAX(10, (int)INTER_BITS);
4166 const int AB_SCALE = 1 << AB_BITS;
4168 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
4169 int depth = src.depth();
4170 int channels = src.channels();
4171 if( ( depth == CV_8U || depth == CV_16U || depth == CV_32F ) &&
4172 ( channels == 1 || channels == 3 || channels == 4 ) &&
4173 ( borderType == cv::BORDER_TRANSPARENT || ( borderType == cv::BORDER_CONSTANT ) ) )
4175 int type = src.type();
4176 ippiWarpAffineBackFunc ippFunc =
4177 type == CV_8UC1 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_8u_C1R :
4178 type == CV_8UC3 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_8u_C3R :
4179 type == CV_8UC4 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_8u_C4R :
4180 type == CV_16UC1 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_16u_C1R :
4181 type == CV_16UC3 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_16u_C3R :
4182 type == CV_16UC4 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_16u_C4R :
4183 type == CV_32FC1 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_32f_C1R :
4184 type == CV_32FC3 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_32f_C3R :
4185 type == CV_32FC4 ? (ippiWarpAffineBackFunc)ippiWarpAffineBack_32f_C4R :
4188 flags == INTER_LINEAR ? IPPI_INTER_LINEAR :
4189 flags == INTER_NEAREST ? IPPI_INTER_NN :
4190 flags == INTER_CUBIC ? IPPI_INTER_CUBIC :
4192 if( mode && ippFunc )
4194 double coeffs[2][3];
4195 for( int i = 0; i < 2; i++ )
4197 for( int j = 0; j < 3; j++ )
4199 coeffs[i][j] = matM.at<double>(i, j);
4203 Range range(0, dst.rows);
4204 IPPwarpAffineInvoker invoker(src, dst, coeffs, mode, borderType, borderValue, ippFunc, &ok);
4205 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
4212 for( x = 0; x < dst.cols; x++ )
4214 adelta[x] = saturate_cast<int>(M[0]*x*AB_SCALE);
4215 bdelta[x] = saturate_cast<int>(M[3]*x*AB_SCALE);
4218 Range range(0, dst.rows);
4219 warpAffineInvoker invoker(src, dst, interpolation, borderType,
4220 borderValue, adelta, bdelta, M);
4221 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
4228 class warpPerspectiveInvoker :
4229 public ParallelLoopBody
4233 warpPerspectiveInvoker(const Mat &_src, Mat &_dst, double *_M, int _interpolation,
4234 int _borderType, const Scalar &_borderValue) :
4235 ParallelLoopBody(), src(_src), dst(_dst), M(_M), interpolation(_interpolation),
4236 borderType(_borderType), borderValue(_borderValue)
4240 virtual void operator() (const Range& range) const
4242 const int BLOCK_SZ = 32;
4243 short XY[BLOCK_SZ*BLOCK_SZ*2], A[BLOCK_SZ*BLOCK_SZ];
4244 int x, y, x1, y1, width = dst.cols, height = dst.rows;
4246 int bh0 = std::min(BLOCK_SZ/2, height);
4247 int bw0 = std::min(BLOCK_SZ*BLOCK_SZ/bh0, width);
4248 bh0 = std::min(BLOCK_SZ*BLOCK_SZ/bw0, height);
4250 for( y = range.start; y < range.end; y += bh0 )
4252 for( x = 0; x < width; x += bw0 )
4254 int bw = std::min( bw0, width - x);
4255 int bh = std::min( bh0, range.end - y); // height
4257 Mat _XY(bh, bw, CV_16SC2, XY), matA;
4258 Mat dpart(dst, Rect(x, y, bw, bh));
4260 for( y1 = 0; y1 < bh; y1++ )
4262 short* xy = XY + y1*bw*2;
4263 double X0 = M[0]*x + M[1]*(y + y1) + M[2];
4264 double Y0 = M[3]*x + M[4]*(y + y1) + M[5];
4265 double W0 = M[6]*x + M[7]*(y + y1) + M[8];
4267 if( interpolation == INTER_NEAREST )
4268 for( x1 = 0; x1 < bw; x1++ )
4270 double W = W0 + M[6]*x1;
4272 double fX = std::max((double)INT_MIN, std::min((double)INT_MAX, (X0 + M[0]*x1)*W));
4273 double fY = std::max((double)INT_MIN, std::min((double)INT_MAX, (Y0 + M[3]*x1)*W));
4274 int X = saturate_cast<int>(fX);
4275 int Y = saturate_cast<int>(fY);
4277 xy[x1*2] = saturate_cast<short>(X);
4278 xy[x1*2+1] = saturate_cast<short>(Y);
4282 short* alpha = A + y1*bw;
4283 for( x1 = 0; x1 < bw; x1++ )
4285 double W = W0 + M[6]*x1;
4286 W = W ? INTER_TAB_SIZE/W : 0;
4287 double fX = std::max((double)INT_MIN, std::min((double)INT_MAX, (X0 + M[0]*x1)*W));
4288 double fY = std::max((double)INT_MIN, std::min((double)INT_MAX, (Y0 + M[3]*x1)*W));
4289 int X = saturate_cast<int>(fX);
4290 int Y = saturate_cast<int>(fY);
4292 xy[x1*2] = saturate_cast<short>(X >> INTER_BITS);
4293 xy[x1*2+1] = saturate_cast<short>(Y >> INTER_BITS);
4294 alpha[x1] = (short)((Y & (INTER_TAB_SIZE-1))*INTER_TAB_SIZE +
4295 (X & (INTER_TAB_SIZE-1)));
4300 if( interpolation == INTER_NEAREST )
4301 remap( src, dpart, _XY, Mat(), interpolation, borderType, borderValue );
4304 Mat _matA(bh, bw, CV_16U, A);
4305 remap( src, dpart, _XY, _matA, interpolation, borderType, borderValue );
4315 int interpolation, borderType;
4319 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
4320 class IPPwarpPerspectiveInvoker :
4321 public ParallelLoopBody
4324 IPPwarpPerspectiveInvoker(Mat &_src, Mat &_dst, double (&_coeffs)[3][3], int &_interpolation, int &_borderType, const Scalar &_borderValue, ippiWarpPerspectiveBackFunc _func, bool *_ok) :
4325 ParallelLoopBody(), src(_src), dst(_dst), mode(_interpolation), coeffs(_coeffs), borderType(_borderType), borderValue(_borderValue), func(_func), ok(_ok)
4330 virtual void operator() (const Range& range) const
4332 IppiSize srcsize = {src.cols, src.rows};
4333 IppiRect srcroi = {0, 0, src.cols, src.rows};
4334 IppiRect dstroi = {0, range.start, dst.cols, range.end - range.start};
4335 int cnn = src.channels();
4337 if( borderType == BORDER_CONSTANT )
4339 IppiSize setSize = {dst.cols, range.end - range.start};
4340 void *dataPointer = dst.data + dst.step[0] * range.start;
4341 if( !IPPSet( borderValue, dataPointer, (int)dst.step[0], setSize, cnn, src.depth() ) )
4347 if( func(src.data, srcsize, (int)src.step[0], srcroi, dst.data, (int)dst.step[0], dstroi, coeffs, mode) < 0)
4353 double (&coeffs)[3][3];
4356 const Scalar borderValue;
4357 ippiWarpPerspectiveBackFunc func;
4359 const IPPwarpPerspectiveInvoker& operator= (const IPPwarpPerspectiveInvoker&);
4365 void cv::warpPerspective( InputArray _src, OutputArray _dst, InputArray _M0,
4366 Size dsize, int flags, int borderType, const Scalar& borderValue )
4368 CV_Assert( _src.total() > 0 );
4370 CV_OCL_RUN(_src.dims() <= 2 && _dst.isUMat(),
4371 ocl_warpTransform(_src, _dst, _M0, dsize, flags, borderType, borderValue,
4372 OCL_OP_PERSPECTIVE))
4374 Mat src = _src.getMat(), M0 = _M0.getMat();
4375 _dst.create( dsize.area() == 0 ? src.size() : dsize, src.type() );
4376 Mat dst = _dst.getMat();
4378 if( dst.data == src.data )
4382 Mat matM(3, 3, CV_64F, M);
4383 int interpolation = flags & INTER_MAX;
4384 if( interpolation == INTER_AREA )
4385 interpolation = INTER_LINEAR;
4387 CV_Assert( (M0.type() == CV_32F || M0.type() == CV_64F) && M0.rows == 3 && M0.cols == 3 );
4388 M0.convertTo(matM, matM.type());
4390 #ifdef HAVE_TEGRA_OPTIMIZATION
4391 if( tegra::warpPerspective(src, dst, M, flags, borderType, borderValue) )
4395 if( !(flags & WARP_INVERSE_MAP) )
4398 #if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
4399 int depth = src.depth();
4400 int channels = src.channels();
4401 if( ( depth == CV_8U || depth == CV_16U || depth == CV_32F ) &&
4402 ( channels == 1 || channels == 3 || channels == 4 ) &&
4403 ( borderType == cv::BORDER_TRANSPARENT || borderType == cv::BORDER_CONSTANT ) )
4405 int type = src.type();
4406 ippiWarpPerspectiveBackFunc ippFunc =
4407 type == CV_8UC1 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_8u_C1R :
4408 type == CV_8UC3 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_8u_C3R :
4409 type == CV_8UC4 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_8u_C4R :
4410 type == CV_16UC1 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_16u_C1R :
4411 type == CV_16UC3 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_16u_C3R :
4412 type == CV_16UC4 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_16u_C4R :
4413 type == CV_32FC1 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_32f_C1R :
4414 type == CV_32FC3 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_32f_C3R :
4415 type == CV_32FC4 ? (ippiWarpPerspectiveBackFunc)ippiWarpPerspectiveBack_32f_C4R :
4418 flags == INTER_LINEAR ? IPPI_INTER_LINEAR :
4419 flags == INTER_NEAREST ? IPPI_INTER_NN :
4420 flags == INTER_CUBIC ? IPPI_INTER_CUBIC :
4422 if( mode && ippFunc )
4424 double coeffs[3][3];
4425 for( int i = 0; i < 3; i++ )
4427 for( int j = 0; j < 3; j++ )
4429 coeffs[i][j] = matM.at<double>(i, j);
4433 Range range(0, dst.rows);
4434 IPPwarpPerspectiveInvoker invoker(src, dst, coeffs, mode, borderType, borderValue, ippFunc, &ok);
4435 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
4442 Range range(0, dst.rows);
4443 warpPerspectiveInvoker invoker(src, dst, M, interpolation, borderType, borderValue);
4444 parallel_for_(range, invoker, dst.total()/(double)(1<<16));
4448 cv::Mat cv::getRotationMatrix2D( Point2f center, double angle, double scale )
4451 double alpha = cos(angle)*scale;
4452 double beta = sin(angle)*scale;
4454 Mat M(2, 3, CV_64F);
4455 double* m = (double*)M.data;
4459 m[2] = (1-alpha)*center.x - beta*center.y;
4462 m[5] = beta*center.x + (1-alpha)*center.y;
4467 /* Calculates coefficients of perspective transformation
4468 * which maps (xi,yi) to (ui,vi), (i=1,2,3,4):
4470 * c00*xi + c01*yi + c02
4471 * ui = ---------------------
4472 * c20*xi + c21*yi + c22
4474 * c10*xi + c11*yi + c12
4475 * vi = ---------------------
4476 * c20*xi + c21*yi + c22
4478 * Coefficients are calculated by solving linear system:
4479 * / x0 y0 1 0 0 0 -x0*u0 -y0*u0 \ /c00\ /u0\
4480 * | x1 y1 1 0 0 0 -x1*u1 -y1*u1 | |c01| |u1|
4481 * | x2 y2 1 0 0 0 -x2*u2 -y2*u2 | |c02| |u2|
4482 * | x3 y3 1 0 0 0 -x3*u3 -y3*u3 |.|c10|=|u3|,
4483 * | 0 0 0 x0 y0 1 -x0*v0 -y0*v0 | |c11| |v0|
4484 * | 0 0 0 x1 y1 1 -x1*v1 -y1*v1 | |c12| |v1|
4485 * | 0 0 0 x2 y2 1 -x2*v2 -y2*v2 | |c20| |v2|
4486 * \ 0 0 0 x3 y3 1 -x3*v3 -y3*v3 / \c21/ \v3/
4489 * cij - matrix coefficients, c22 = 1
4491 cv::Mat cv::getPerspectiveTransform( const Point2f src[], const Point2f dst[] )
4493 Mat M(3, 3, CV_64F), X(8, 1, CV_64F, M.data);
4494 double a[8][8], b[8];
4495 Mat A(8, 8, CV_64F, a), B(8, 1, CV_64F, b);
4497 for( int i = 0; i < 4; ++i )
4499 a[i][0] = a[i+4][3] = src[i].x;
4500 a[i][1] = a[i+4][4] = src[i].y;
4501 a[i][2] = a[i+4][5] = 1;
4502 a[i][3] = a[i][4] = a[i][5] =
4503 a[i+4][0] = a[i+4][1] = a[i+4][2] = 0;
4504 a[i][6] = -src[i].x*dst[i].x;
4505 a[i][7] = -src[i].y*dst[i].x;
4506 a[i+4][6] = -src[i].x*dst[i].y;
4507 a[i+4][7] = -src[i].y*dst[i].y;
4512 solve( A, B, X, DECOMP_SVD );
4513 ((double*)M.data)[8] = 1.;
4518 /* Calculates coefficients of affine transformation
4519 * which maps (xi,yi) to (ui,vi), (i=1,2,3):
4521 * ui = c00*xi + c01*yi + c02
4523 * vi = c10*xi + c11*yi + c12
4525 * Coefficients are calculated by solving linear system:
4526 * / x0 y0 1 0 0 0 \ /c00\ /u0\
4527 * | x1 y1 1 0 0 0 | |c01| |u1|
4528 * | x2 y2 1 0 0 0 | |c02| |u2|
4529 * | 0 0 0 x0 y0 1 | |c10| |v0|
4530 * | 0 0 0 x1 y1 1 | |c11| |v1|
4531 * \ 0 0 0 x2 y2 1 / |c12| |v2|
4534 * cij - matrix coefficients
4537 cv::Mat cv::getAffineTransform( const Point2f src[], const Point2f dst[] )
4539 Mat M(2, 3, CV_64F), X(6, 1, CV_64F, M.data);
4540 double a[6*6], b[6];
4541 Mat A(6, 6, CV_64F, a), B(6, 1, CV_64F, b);
4543 for( int i = 0; i < 3; i++ )
4547 a[j] = a[k+3] = src[i].x;
4548 a[j+1] = a[k+4] = src[i].y;
4549 a[j+2] = a[k+5] = 1;
4550 a[j+3] = a[j+4] = a[j+5] = 0;
4551 a[k] = a[k+1] = a[k+2] = 0;
4553 b[i*2+1] = dst[i].y;
4560 void cv::invertAffineTransform(InputArray _matM, OutputArray __iM)
4562 Mat matM = _matM.getMat();
4563 CV_Assert(matM.rows == 2 && matM.cols == 3);
4564 __iM.create(2, 3, matM.type());
4565 Mat _iM = __iM.getMat();
4567 if( matM.type() == CV_32F )
4569 const float* M = (const float*)matM.data;
4570 float* iM = (float*)_iM.data;
4571 int step = (int)(matM.step/sizeof(M[0])), istep = (int)(_iM.step/sizeof(iM[0]));
4573 double D = M[0]*M[step+1] - M[1]*M[step];
4574 D = D != 0 ? 1./D : 0;
4575 double A11 = M[step+1]*D, A22 = M[0]*D, A12 = -M[1]*D, A21 = -M[step]*D;
4576 double b1 = -A11*M[2] - A12*M[step+2];
4577 double b2 = -A21*M[2] - A22*M[step+2];
4579 iM[0] = (float)A11; iM[1] = (float)A12; iM[2] = (float)b1;
4580 iM[istep] = (float)A21; iM[istep+1] = (float)A22; iM[istep+2] = (float)b2;
4582 else if( matM.type() == CV_64F )
4584 const double* M = (const double*)matM.data;
4585 double* iM = (double*)_iM.data;
4586 int step = (int)(matM.step/sizeof(M[0])), istep = (int)(_iM.step/sizeof(iM[0]));
4588 double D = M[0]*M[step+1] - M[1]*M[step];
4589 D = D != 0 ? 1./D : 0;
4590 double A11 = M[step+1]*D, A22 = M[0]*D, A12 = -M[1]*D, A21 = -M[step]*D;
4591 double b1 = -A11*M[2] - A12*M[step+2];
4592 double b2 = -A21*M[2] - A22*M[step+2];
4594 iM[0] = A11; iM[1] = A12; iM[2] = b1;
4595 iM[istep] = A21; iM[istep+1] = A22; iM[istep+2] = b2;
4598 CV_Error( CV_StsUnsupportedFormat, "" );
4601 cv::Mat cv::getPerspectiveTransform(InputArray _src, InputArray _dst)
4603 Mat src = _src.getMat(), dst = _dst.getMat();
4604 CV_Assert(src.checkVector(2, CV_32F) == 4 && dst.checkVector(2, CV_32F) == 4);
4605 return getPerspectiveTransform((const Point2f*)src.data, (const Point2f*)dst.data);
4608 cv::Mat cv::getAffineTransform(InputArray _src, InputArray _dst)
4610 Mat src = _src.getMat(), dst = _dst.getMat();
4611 CV_Assert(src.checkVector(2, CV_32F) == 3 && dst.checkVector(2, CV_32F) == 3);
4612 return getAffineTransform((const Point2f*)src.data, (const Point2f*)dst.data);
4616 cvResize( const CvArr* srcarr, CvArr* dstarr, int method )
4618 cv::Mat src = cv::cvarrToMat(srcarr), dst = cv::cvarrToMat(dstarr);
4619 CV_Assert( src.type() == dst.type() );
4620 cv::resize( src, dst, dst.size(), (double)dst.cols/src.cols,
4621 (double)dst.rows/src.rows, method );
4626 cvWarpAffine( const CvArr* srcarr, CvArr* dstarr, const CvMat* marr,
4627 int flags, CvScalar fillval )
4629 cv::Mat src = cv::cvarrToMat(srcarr), dst = cv::cvarrToMat(dstarr);
4630 cv::Mat matrix = cv::cvarrToMat(marr);
4631 CV_Assert( src.type() == dst.type() );
4632 cv::warpAffine( src, dst, matrix, dst.size(), flags,
4633 (flags & CV_WARP_FILL_OUTLIERS) ? cv::BORDER_CONSTANT : cv::BORDER_TRANSPARENT,
4638 cvWarpPerspective( const CvArr* srcarr, CvArr* dstarr, const CvMat* marr,
4639 int flags, CvScalar fillval )
4641 cv::Mat src = cv::cvarrToMat(srcarr), dst = cv::cvarrToMat(dstarr);
4642 cv::Mat matrix = cv::cvarrToMat(marr);
4643 CV_Assert( src.type() == dst.type() );
4644 cv::warpPerspective( src, dst, matrix, dst.size(), flags,
4645 (flags & CV_WARP_FILL_OUTLIERS) ? cv::BORDER_CONSTANT : cv::BORDER_TRANSPARENT,
4650 cvRemap( const CvArr* srcarr, CvArr* dstarr,
4651 const CvArr* _mapx, const CvArr* _mapy,
4652 int flags, CvScalar fillval )
4654 cv::Mat src = cv::cvarrToMat(srcarr), dst = cv::cvarrToMat(dstarr), dst0 = dst;
4655 cv::Mat mapx = cv::cvarrToMat(_mapx), mapy = cv::cvarrToMat(_mapy);
4656 CV_Assert( src.type() == dst.type() && dst.size() == mapx.size() );
4657 cv::remap( src, dst, mapx, mapy, flags & cv::INTER_MAX,
4658 (flags & CV_WARP_FILL_OUTLIERS) ? cv::BORDER_CONSTANT : cv::BORDER_TRANSPARENT,
4660 CV_Assert( dst0.data == dst.data );
4665 cv2DRotationMatrix( CvPoint2D32f center, double angle,
4666 double scale, CvMat* matrix )
4668 cv::Mat M0 = cv::cvarrToMat(matrix), M = cv::getRotationMatrix2D(center, angle, scale);
4669 CV_Assert( M.size() == M0.size() );
4670 M.convertTo(M0, M0.type());
4676 cvGetPerspectiveTransform( const CvPoint2D32f* src,
4677 const CvPoint2D32f* dst,
4680 cv::Mat M0 = cv::cvarrToMat(matrix),
4681 M = cv::getPerspectiveTransform((const cv::Point2f*)src, (const cv::Point2f*)dst);
4682 CV_Assert( M.size() == M0.size() );
4683 M.convertTo(M0, M0.type());
4689 cvGetAffineTransform( const CvPoint2D32f* src,
4690 const CvPoint2D32f* dst,
4693 cv::Mat M0 = cv::cvarrToMat(matrix),
4694 M = cv::getAffineTransform((const cv::Point2f*)src, (const cv::Point2f*)dst);
4695 CV_Assert( M.size() == M0.size() );
4696 M.convertTo(M0, M0.type());
4702 cvConvertMaps( const CvArr* arr1, const CvArr* arr2, CvArr* dstarr1, CvArr* dstarr2 )
4704 cv::Mat map1 = cv::cvarrToMat(arr1), map2;
4705 cv::Mat dstmap1 = cv::cvarrToMat(dstarr1), dstmap2;
4708 map2 = cv::cvarrToMat(arr2);
4711 dstmap2 = cv::cvarrToMat(dstarr2);
4712 if( dstmap2.type() == CV_16SC1 )
4713 dstmap2 = cv::Mat(dstmap2.size(), CV_16UC1, dstmap2.data, dstmap2.step);
4716 cv::convertMaps( map1, map2, dstmap1, dstmap2, dstmap1.type(), false );
4719 /****************************************************************************************\
4720 * Log-Polar Transform *
4721 \****************************************************************************************/
4723 /* now it is done via Remap; more correct implementation should use
4724 some super-sampling technique outside of the "fovea" circle */
4726 cvLogPolar( const CvArr* srcarr, CvArr* dstarr,
4727 CvPoint2D32f center, double M, int flags )
4729 cv::Ptr<CvMat> mapx, mapy;
4731 CvMat srcstub, *src = cvGetMat(srcarr, &srcstub);
4732 CvMat dststub, *dst = cvGetMat(dstarr, &dststub);
4733 CvSize ssize, dsize;
4735 if( !CV_ARE_TYPES_EQ( src, dst ))
4736 CV_Error( CV_StsUnmatchedFormats, "" );
4739 CV_Error( CV_StsOutOfRange, "M should be >0" );
4741 ssize = cvGetMatSize(src);
4742 dsize = cvGetMatSize(dst);
4744 mapx.reset(cvCreateMat( dsize.height, dsize.width, CV_32F ));
4745 mapy.reset(cvCreateMat( dsize.height, dsize.width, CV_32F ));
4747 if( !(flags & CV_WARP_INVERSE_MAP) )
4750 cv::AutoBuffer<double> _exp_tab(dsize.width);
4751 double* exp_tab = _exp_tab;
4753 for( rho = 0; rho < dst->width; rho++ )
4754 exp_tab[rho] = std::exp(rho/M);
4756 for( phi = 0; phi < dsize.height; phi++ )
4758 double cp = cos(phi*2*CV_PI/dsize.height);
4759 double sp = sin(phi*2*CV_PI/dsize.height);
4760 float* mx = (float*)(mapx->data.ptr + phi*mapx->step);
4761 float* my = (float*)(mapy->data.ptr + phi*mapy->step);
4763 for( rho = 0; rho < dsize.width; rho++ )
4765 double r = exp_tab[rho];
4766 double x = r*cp + center.x;
4767 double y = r*sp + center.y;
4777 CvMat bufx, bufy, bufp, bufa;
4778 double ascale = ssize.height/(2*CV_PI);
4779 cv::AutoBuffer<float> _buf(4*dsize.width);
4782 bufx = cvMat( 1, dsize.width, CV_32F, buf );
4783 bufy = cvMat( 1, dsize.width, CV_32F, buf + dsize.width );
4784 bufp = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*2 );
4785 bufa = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*3 );
4787 for( x = 0; x < dsize.width; x++ )
4788 bufx.data.fl[x] = (float)x - center.x;
4790 for( y = 0; y < dsize.height; y++ )
4792 float* mx = (float*)(mapx->data.ptr + y*mapx->step);
4793 float* my = (float*)(mapy->data.ptr + y*mapy->step);
4795 for( x = 0; x < dsize.width; x++ )
4796 bufy.data.fl[x] = (float)y - center.y;
4799 cvCartToPolar( &bufx, &bufy, &bufp, &bufa );
4801 for( x = 0; x < dsize.width; x++ )
4802 bufp.data.fl[x] += 1.f;
4804 cvLog( &bufp, &bufp );
4806 for( x = 0; x < dsize.width; x++ )
4808 double rho = bufp.data.fl[x]*M;
4809 double phi = bufa.data.fl[x]*ascale;
4815 for( x = 0; x < dsize.width; x++ )
4817 double xx = bufx.data.fl[x];
4818 double yy = bufy.data.fl[x];
4820 double p = log(std::sqrt(xx*xx + yy*yy) + 1.)*M;
4821 double a = atan2(yy,xx);
4833 cvRemap( src, dst, mapx, mapy, flags, cvScalarAll(0) );
4836 void cv::logPolar( InputArray _src, OutputArray _dst,
4837 Point2f center, double M, int flags )
4839 Mat src = _src.getMat();
4840 _dst.create( src.size(), src.type() );
4841 CvMat c_src = src, c_dst = _dst.getMat();
4842 cvLogPolar( &c_src, &c_dst, center, M, flags );
4845 /****************************************************************************************
4846 Linear-Polar Transform
4847 J.L. Blanco, Apr 2009
4848 ****************************************************************************************/
4850 void cvLinearPolar( const CvArr* srcarr, CvArr* dstarr,
4851 CvPoint2D32f center, double maxRadius, int flags )
4853 cv::Ptr<CvMat> mapx, mapy;
4855 CvMat srcstub, *src = (CvMat*)srcarr;
4856 CvMat dststub, *dst = (CvMat*)dstarr;
4857 CvSize ssize, dsize;
4859 src = cvGetMat( srcarr, &srcstub,0,0 );
4860 dst = cvGetMat( dstarr, &dststub,0,0 );
4862 if( !CV_ARE_TYPES_EQ( src, dst ))
4863 CV_Error( CV_StsUnmatchedFormats, "" );
4865 ssize.width = src->cols;
4866 ssize.height = src->rows;
4867 dsize.width = dst->cols;
4868 dsize.height = dst->rows;
4870 mapx.reset(cvCreateMat( dsize.height, dsize.width, CV_32F ));
4871 mapy.reset(cvCreateMat( dsize.height, dsize.width, CV_32F ));
4873 if( !(flags & CV_WARP_INVERSE_MAP) )
4877 for( phi = 0; phi < dsize.height; phi++ )
4879 double cp = cos(phi*2*CV_PI/dsize.height);
4880 double sp = sin(phi*2*CV_PI/dsize.height);
4881 float* mx = (float*)(mapx->data.ptr + phi*mapx->step);
4882 float* my = (float*)(mapy->data.ptr + phi*mapy->step);
4884 for( rho = 0; rho < dsize.width; rho++ )
4886 double r = maxRadius*(rho+1)/dsize.width;
4887 double x = r*cp + center.x;
4888 double y = r*sp + center.y;
4898 CvMat bufx, bufy, bufp, bufa;
4899 const double ascale = ssize.height/(2*CV_PI);
4900 const double pscale = ssize.width/maxRadius;
4902 cv::AutoBuffer<float> _buf(4*dsize.width);
4905 bufx = cvMat( 1, dsize.width, CV_32F, buf );
4906 bufy = cvMat( 1, dsize.width, CV_32F, buf + dsize.width );
4907 bufp = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*2 );
4908 bufa = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*3 );
4910 for( x = 0; x < dsize.width; x++ )
4911 bufx.data.fl[x] = (float)x - center.x;
4913 for( y = 0; y < dsize.height; y++ )
4915 float* mx = (float*)(mapx->data.ptr + y*mapx->step);
4916 float* my = (float*)(mapy->data.ptr + y*mapy->step);
4918 for( x = 0; x < dsize.width; x++ )
4919 bufy.data.fl[x] = (float)y - center.y;
4921 cvCartToPolar( &bufx, &bufy, &bufp, &bufa, 0 );
4923 for( x = 0; x < dsize.width; x++ )
4924 bufp.data.fl[x] += 1.f;
4926 for( x = 0; x < dsize.width; x++ )
4928 double rho = bufp.data.fl[x]*pscale;
4929 double phi = bufa.data.fl[x]*ascale;
4936 cvRemap( src, dst, mapx, mapy, flags, cvScalarAll(0) );
4939 void cv::linearPolar( InputArray _src, OutputArray _dst,
4940 Point2f center, double maxRadius, int flags )
4942 Mat src = _src.getMat();
4943 _dst.create( src.size(), src.type() );
4944 CvMat c_src = src, c_dst = _dst.getMat();
4945 cvLinearPolar( &c_src, &c_dst, center, maxRadius, flags );