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 #include "precomp.hpp"
46 #if defined _M_IX86 && defined _MSC_VER && _MSC_VER < 1700
47 #pragma float_control(precise, on)
53 /****************************************************************************************\
54 * LU & Cholesky implementation for small matrices *
55 \****************************************************************************************/
57 template<typename _Tp> static inline int
58 LUImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n)
61 astep /= sizeof(A[0]);
62 bstep /= sizeof(b[0]);
64 for( i = 0; i < m; i++ )
68 for( j = i+1; j < m; j++ )
69 if( std::abs(A[j*astep + i]) > std::abs(A[k*astep + i]) )
72 if( std::abs(A[k*astep + i]) < std::numeric_limits<_Tp>::epsilon() )
77 for( j = i; j < m; j++ )
78 std::swap(A[i*astep + j], A[k*astep + j]);
80 for( j = 0; j < n; j++ )
81 std::swap(b[i*bstep + j], b[k*bstep + j]);
85 _Tp d = -1/A[i*astep + i];
87 for( j = i+1; j < m; j++ )
89 _Tp alpha = A[j*astep + i]*d;
91 for( k = i+1; k < m; k++ )
92 A[j*astep + k] += alpha*A[i*astep + k];
95 for( k = 0; k < n; k++ )
96 b[j*bstep + k] += alpha*b[i*bstep + k];
104 for( i = m-1; i >= 0; i-- )
105 for( j = 0; j < n; j++ )
107 _Tp s = b[i*bstep + j];
108 for( k = i+1; k < m; k++ )
109 s -= A[i*astep + k]*b[k*bstep + j];
110 b[i*bstep + j] = s*A[i*astep + i];
118 int LU(float* A, size_t astep, int m, float* b, size_t bstep, int n)
120 return LUImpl(A, astep, m, b, bstep, n);
124 int LU(double* A, size_t astep, int m, double* b, size_t bstep, int n)
126 return LUImpl(A, astep, m, b, bstep, n);
130 template<typename _Tp> static inline bool
131 CholImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n)
136 astep /= sizeof(A[0]);
137 bstep /= sizeof(b[0]);
139 for( i = 0; i < m; i++ )
141 for( j = 0; j < i; j++ )
144 for( k = 0; k < j; k++ )
145 s -= L[i*astep + k]*L[j*astep + k];
146 L[i*astep + j] = (_Tp)(s*L[j*astep + j]);
149 for( k = 0; k < j; k++ )
151 double t = L[i*astep + k];
154 if( s < std::numeric_limits<_Tp>::epsilon() )
156 L[i*astep + i] = (_Tp)(1./std::sqrt(s));
169 [ L20 L21 L22 ] y2 b2
170 [ L30 L31 L32 L33 ] y3 b3
172 [ L00 L10 L20 L30 ] x0 y0
173 [ L11 L21 L31 ] x1 = y1
178 for( i = 0; i < m; i++ )
180 for( j = 0; j < n; j++ )
183 for( k = 0; k < i; k++ )
184 s -= L[i*astep + k]*b[k*bstep + j];
185 b[i*bstep + j] = (_Tp)(s*L[i*astep + i]);
189 for( i = m-1; i >= 0; i-- )
191 for( j = 0; j < n; j++ )
194 for( k = m-1; k > i; k-- )
195 s -= L[k*astep + i]*b[k*bstep + j];
196 b[i*bstep + j] = (_Tp)(s*L[i*astep + i]);
204 bool Cholesky(float* A, size_t astep, int m, float* b, size_t bstep, int n)
206 return CholImpl(A, astep, m, b, bstep, n);
209 bool Cholesky(double* A, size_t astep, int m, double* b, size_t bstep, int n)
211 return CholImpl(A, astep, m, b, bstep, n);
215 template<typename _Tp> static inline _Tp hypot(_Tp a, _Tp b)
222 return a*std::sqrt(1 + b*b);
227 return b*std::sqrt(1 + a*a);
233 template<typename _Tp> bool
234 JacobiImpl_( _Tp* A, size_t astep, _Tp* W, _Tp* V, size_t vstep, int n, uchar* buf )
236 const _Tp eps = std::numeric_limits<_Tp>::epsilon();
239 astep /= sizeof(A[0]);
242 vstep /= sizeof(V[0]);
243 for( i = 0; i < n; i++ )
245 for( j = 0; j < n; j++ )
246 V[i*vstep + j] = (_Tp)0;
247 V[i*vstep + i] = (_Tp)1;
251 int iters, maxIters = n*n*30;
253 int* indR = (int*)alignPtr(buf, sizeof(int));
254 int* indC = indR + n;
257 for( k = 0; k < n; k++ )
259 W[k] = A[(astep + 1)*k];
262 for( m = k+1, mv = std::abs(A[astep*k + m]), i = k+2; i < n; i++ )
264 _Tp val = std::abs(A[astep*k+i]);
272 for( m = 0, mv = std::abs(A[k]), i = 1; i < k; i++ )
274 _Tp val = std::abs(A[astep*i+k]);
282 if( n > 1 ) for( iters = 0; iters < maxIters; iters++ )
284 // find index (k,l) of pivot p
285 for( k = 0, mv = std::abs(A[indR[0]]), i = 1; i < n-1; i++ )
287 _Tp val = std::abs(A[astep*i + indR[i]]);
292 for( i = 1; i < n; i++ )
294 _Tp val = std::abs(A[astep*indC[i] + i]);
296 mv = val, k = indC[i], l = i;
299 _Tp p = A[astep*k + l];
300 if( std::abs(p) <= eps )
302 _Tp y = (_Tp)((W[l] - W[k])*0.5);
303 _Tp t = std::abs(y) + hypot(p, y);
306 s = p/s; t = (p/t)*p;
317 #define rotate(v0, v1) a0 = v0, b0 = v1, v0 = a0*c - b0*s, v1 = a0*s + b0*c
319 // rotate rows and columns k and l
320 for( i = 0; i < k; i++ )
321 rotate(A[astep*i+k], A[astep*i+l]);
322 for( i = k+1; i < l; i++ )
323 rotate(A[astep*k+i], A[astep*i+l]);
324 for( i = l+1; i < n; i++ )
325 rotate(A[astep*k+i], A[astep*l+i]);
327 // rotate eigenvectors
329 for( i = 0; i < n; i++ )
330 rotate(V[vstep*k+i], V[vstep*l+i]);
334 for( j = 0; j < 2; j++ )
336 int idx = j == 0 ? k : l;
339 for( m = idx+1, mv = std::abs(A[astep*idx + m]), i = idx+2; i < n; i++ )
341 _Tp val = std::abs(A[astep*idx+i]);
349 for( m = 0, mv = std::abs(A[idx]), i = 1; i < idx; i++ )
351 _Tp val = std::abs(A[astep*i+idx]);
360 // sort eigenvalues & eigenvectors
361 for( k = 0; k < n-1; k++ )
364 for( i = k+1; i < n; i++ )
371 std::swap(W[m], W[k]);
373 for( i = 0; i < n; i++ )
374 std::swap(V[vstep*m + i], V[vstep*k + i]);
381 static bool Jacobi( float* S, size_t sstep, float* e, float* E, size_t estep, int n, uchar* buf )
383 return JacobiImpl_(S, sstep, e, E, estep, n, buf);
386 static bool Jacobi( double* S, size_t sstep, double* e, double* E, size_t estep, int n, uchar* buf )
388 return JacobiImpl_(S, sstep, e, E, estep, n, buf);
392 template<typename T> struct VBLAS
394 int dot(const T*, const T*, int, T*) const { return 0; }
395 int givens(T*, T*, int, T, T) const { return 0; }
396 int givensx(T*, T*, int, T, T, T*, T*) const { return 0; }
400 template<> inline int VBLAS<float>::dot(const float* a, const float* b, int n, float* result) const
405 __m128 s0 = _mm_setzero_ps(), s1 = _mm_setzero_ps();
406 for( ; k <= n - 8; k += 8 )
408 __m128 a0 = _mm_load_ps(a + k), a1 = _mm_load_ps(a + k + 4);
409 __m128 b0 = _mm_load_ps(b + k), b1 = _mm_load_ps(b + k + 4);
411 s0 = _mm_add_ps(s0, _mm_mul_ps(a0, b0));
412 s1 = _mm_add_ps(s1, _mm_mul_ps(a1, b1));
414 s0 = _mm_add_ps(s0, s1);
416 _mm_storeu_ps(sbuf, s0);
417 *result = sbuf[0] + sbuf[1] + sbuf[2] + sbuf[3];
422 template<> inline int VBLAS<float>::givens(float* a, float* b, int n, float c, float s) const
427 __m128 c4 = _mm_set1_ps(c), s4 = _mm_set1_ps(s);
428 for( ; k <= n - 4; k += 4 )
430 __m128 a0 = _mm_load_ps(a + k);
431 __m128 b0 = _mm_load_ps(b + k);
432 __m128 t0 = _mm_add_ps(_mm_mul_ps(a0, c4), _mm_mul_ps(b0, s4));
433 __m128 t1 = _mm_sub_ps(_mm_mul_ps(b0, c4), _mm_mul_ps(a0, s4));
434 _mm_store_ps(a + k, t0);
435 _mm_store_ps(b + k, t1);
441 template<> inline int VBLAS<float>::givensx(float* a, float* b, int n, float c, float s,
442 float* anorm, float* bnorm) const
447 __m128 c4 = _mm_set1_ps(c), s4 = _mm_set1_ps(s);
448 __m128 sa = _mm_setzero_ps(), sb = _mm_setzero_ps();
449 for( ; k <= n - 4; k += 4 )
451 __m128 a0 = _mm_load_ps(a + k);
452 __m128 b0 = _mm_load_ps(b + k);
453 __m128 t0 = _mm_add_ps(_mm_mul_ps(a0, c4), _mm_mul_ps(b0, s4));
454 __m128 t1 = _mm_sub_ps(_mm_mul_ps(b0, c4), _mm_mul_ps(a0, s4));
455 _mm_store_ps(a + k, t0);
456 _mm_store_ps(b + k, t1);
457 sa = _mm_add_ps(sa, _mm_mul_ps(t0, t0));
458 sb = _mm_add_ps(sb, _mm_mul_ps(t1, t1));
460 float abuf[4], bbuf[4];
461 _mm_storeu_ps(abuf, sa);
462 _mm_storeu_ps(bbuf, sb);
463 *anorm = abuf[0] + abuf[1] + abuf[2] + abuf[3];
464 *bnorm = bbuf[0] + bbuf[1] + bbuf[2] + bbuf[3];
469 template<> inline int VBLAS<double>::dot(const double* a, const double* b, int n, double* result) const
474 __m128d s0 = _mm_setzero_pd(), s1 = _mm_setzero_pd();
475 for( ; k <= n - 4; k += 4 )
477 __m128d a0 = _mm_load_pd(a + k), a1 = _mm_load_pd(a + k + 2);
478 __m128d b0 = _mm_load_pd(b + k), b1 = _mm_load_pd(b + k + 2);
480 s0 = _mm_add_pd(s0, _mm_mul_pd(a0, b0));
481 s1 = _mm_add_pd(s1, _mm_mul_pd(a1, b1));
483 s0 = _mm_add_pd(s0, s1);
485 _mm_storeu_pd(sbuf, s0);
486 *result = sbuf[0] + sbuf[1];
491 template<> inline int VBLAS<double>::givens(double* a, double* b, int n, double c, double s) const
494 __m128d c2 = _mm_set1_pd(c), s2 = _mm_set1_pd(s);
495 for( ; k <= n - 2; k += 2 )
497 __m128d a0 = _mm_load_pd(a + k);
498 __m128d b0 = _mm_load_pd(b + k);
499 __m128d t0 = _mm_add_pd(_mm_mul_pd(a0, c2), _mm_mul_pd(b0, s2));
500 __m128d t1 = _mm_sub_pd(_mm_mul_pd(b0, c2), _mm_mul_pd(a0, s2));
501 _mm_store_pd(a + k, t0);
502 _mm_store_pd(b + k, t1);
508 template<> inline int VBLAS<double>::givensx(double* a, double* b, int n, double c, double s,
509 double* anorm, double* bnorm) const
512 __m128d c2 = _mm_set1_pd(c), s2 = _mm_set1_pd(s);
513 __m128d sa = _mm_setzero_pd(), sb = _mm_setzero_pd();
514 for( ; k <= n - 2; k += 2 )
516 __m128d a0 = _mm_load_pd(a + k);
517 __m128d b0 = _mm_load_pd(b + k);
518 __m128d t0 = _mm_add_pd(_mm_mul_pd(a0, c2), _mm_mul_pd(b0, s2));
519 __m128d t1 = _mm_sub_pd(_mm_mul_pd(b0, c2), _mm_mul_pd(a0, s2));
520 _mm_store_pd(a + k, t0);
521 _mm_store_pd(b + k, t1);
522 sa = _mm_add_pd(sa, _mm_mul_pd(t0, t0));
523 sb = _mm_add_pd(sb, _mm_mul_pd(t1, t1));
525 double abuf[2], bbuf[2];
526 _mm_storeu_pd(abuf, sa);
527 _mm_storeu_pd(bbuf, sb);
528 *anorm = abuf[0] + abuf[1];
529 *bnorm = bbuf[0] + bbuf[1];
534 template<typename _Tp> void
535 JacobiSVDImpl_(_Tp* At, size_t astep, _Tp* _W, _Tp* Vt, size_t vstep,
536 int m, int n, int n1, double minval, _Tp eps)
539 AutoBuffer<double> Wbuf(n);
541 int i, j, k, iter, max_iter = std::max(m, 30);
544 astep /= sizeof(At[0]);
545 vstep /= sizeof(Vt[0]);
547 for( i = 0; i < n; i++ )
549 for( k = 0, sd = 0; k < m; k++ )
551 _Tp t = At[i*astep + k];
558 for( k = 0; k < n; k++ )
564 for( iter = 0; iter < max_iter; iter++ )
566 bool changed = false;
568 for( i = 0; i < n-1; i++ )
569 for( j = i+1; j < n; j++ )
571 _Tp *Ai = At + i*astep, *Aj = At + j*astep;
572 double a = W[i], p = 0, b = W[j];
574 for( k = 0; k < m; k++ )
575 p += (double)Ai[k]*Aj[k];
577 if( std::abs(p) <= eps*std::sqrt((double)a*b) )
581 double beta = a - b, gamma = hypot((double)p, beta);
584 double delta = (gamma - beta)*0.5;
585 s = (_Tp)std::sqrt(delta/gamma);
586 c = (_Tp)(p/(gamma*s*2));
590 c = (_Tp)std::sqrt((gamma + beta)/(gamma*2));
591 s = (_Tp)(p/(gamma*c*2));
595 for( k = 0; k < m; k++ )
597 _Tp t0 = c*Ai[k] + s*Aj[k];
598 _Tp t1 = -s*Ai[k] + c*Aj[k];
599 Ai[k] = t0; Aj[k] = t1;
601 a += (double)t0*t0; b += (double)t1*t1;
609 _Tp *Vi = Vt + i*vstep, *Vj = Vt + j*vstep;
610 k = vblas.givens(Vi, Vj, n, c, s);
614 _Tp t0 = c*Vi[k] + s*Vj[k];
615 _Tp t1 = -s*Vi[k] + c*Vj[k];
616 Vi[k] = t0; Vj[k] = t1;
624 for( i = 0; i < n; i++ )
626 for( k = 0, sd = 0; k < m; k++ )
628 _Tp t = At[i*astep + k];
631 W[i] = std::sqrt(sd);
634 for( i = 0; i < n-1; i++ )
637 for( k = i+1; k < n; k++ )
644 std::swap(W[i], W[j]);
647 for( k = 0; k < m; k++ )
648 std::swap(At[i*astep + k], At[j*astep + k]);
650 for( k = 0; k < n; k++ )
651 std::swap(Vt[i*vstep + k], Vt[j*vstep + k]);
656 for( i = 0; i < n; i++ )
663 for( i = 0; i < n1; i++ )
665 sd = i < n ? W[i] : 0;
667 while( sd <= minval )
669 // if we got a zero singular value, then in order to get the corresponding left singular vector
670 // we generate a random vector, project it to the previously computed left singular vectors,
671 // subtract the projection and normalize the difference.
672 const _Tp val0 = (_Tp)(1./m);
673 for( k = 0; k < m; k++ )
675 _Tp val = (rng.next() & 256) != 0 ? val0 : -val0;
676 At[i*astep + k] = val;
678 for( iter = 0; iter < 2; iter++ )
680 for( j = 0; j < i; j++ )
683 for( k = 0; k < m; k++ )
684 sd += At[i*astep + k]*At[j*astep + k];
686 for( k = 0; k < m; k++ )
688 _Tp t = (_Tp)(At[i*astep + k] - sd*At[j*astep + k]);
692 asum = asum ? 1/asum : 0;
693 for( k = 0; k < m; k++ )
694 At[i*astep + k] *= asum;
698 for( k = 0; k < m; k++ )
700 _Tp t = At[i*astep + k];
707 for( k = 0; k < m; k++ )
708 At[i*astep + k] *= s;
713 static void JacobiSVD(float* At, size_t astep, float* W, float* Vt, size_t vstep, int m, int n, int n1=-1)
715 JacobiSVDImpl_(At, astep, W, Vt, vstep, m, n, !Vt ? 0 : n1 < 0 ? n : n1, FLT_MIN, FLT_EPSILON*2);
718 static void JacobiSVD(double* At, size_t astep, double* W, double* Vt, size_t vstep, int m, int n, int n1=-1)
720 JacobiSVDImpl_(At, astep, W, Vt, vstep, m, n, !Vt ? 0 : n1 < 0 ? n : n1, DBL_MIN, DBL_EPSILON*10);
723 /* y[0:m,0:n] += diag(a[0:1,0:m]) * x[0:m,0:n] */
724 template<typename T1, typename T2, typename T3> static void
725 MatrAXPY( int m, int n, const T1* x, int dx,
726 const T2* a, int inca, T3* y, int dy )
729 for( i = 0; i < m; i++, x += dx, y += dy )
733 #if CV_ENABLE_UNROLLED
734 for(; j <= n - 4; j += 4 )
736 T3 t0 = (T3)(y[j] + s*x[j]);
737 T3 t1 = (T3)(y[j+1] + s*x[j+1]);
740 t0 = (T3)(y[j+2] + s*x[j+2]);
741 t1 = (T3)(y[j+3] + s*x[j+3]);
747 y[j] = (T3)(y[j] + s*x[j]);
751 template<typename T> static void
752 SVBkSbImpl_( int m, int n, const T* w, int incw,
753 const T* u, int ldu, bool uT,
754 const T* v, int ldv, bool vT,
755 const T* b, int ldb, int nb,
756 T* x, int ldx, double* buffer, T eps )
758 double threshold = 0;
759 int udelta0 = uT ? ldu : 1, udelta1 = uT ? 1 : ldu;
760 int vdelta0 = vT ? ldv : 1, vdelta1 = vT ? 1 : ldv;
761 int i, j, nm = std::min(m, n);
766 for( i = 0; i < n; i++ )
767 for( j = 0; j < nb; j++ )
770 for( i = 0; i < nm; i++ )
771 threshold += w[i*incw];
774 // v * inv(w) * uT * b
775 for( i = 0; i < nm; i++, u += udelta0, v += vdelta0 )
777 double wi = w[i*incw];
778 if( (double)std::abs(wi) <= threshold )
786 for( j = 0; j < m; j++ )
787 s += u[j*udelta1]*b[j*ldb];
792 for( j = 0; j < n; j++ )
793 x[j*ldx] = (T)(x[j*ldx] + s*v[j*vdelta1]);
799 for( j = 0; j < nb; j++ )
801 MatrAXPY( m, nb, b, ldb, u, udelta1, buffer, 0 );
802 for( j = 0; j < nb; j++ )
807 for( j = 0; j < nb; j++ )
808 buffer[j] = u[j*udelta1]*wi;
810 MatrAXPY( n, nb, buffer, 0, v, vdelta1, x, ldx );
816 SVBkSb( int m, int n, const float* w, size_t wstep,
817 const float* u, size_t ustep, bool uT,
818 const float* v, size_t vstep, bool vT,
819 const float* b, size_t bstep, int nb,
820 float* x, size_t xstep, uchar* buffer )
822 SVBkSbImpl_(m, n, w, wstep ? (int)(wstep/sizeof(w[0])) : 1,
823 u, (int)(ustep/sizeof(u[0])), uT,
824 v, (int)(vstep/sizeof(v[0])), vT,
825 b, (int)(bstep/sizeof(b[0])), nb,
826 x, (int)(xstep/sizeof(x[0])),
827 (double*)alignPtr(buffer, sizeof(double)), (float)(DBL_EPSILON*2) );
831 SVBkSb( int m, int n, const double* w, size_t wstep,
832 const double* u, size_t ustep, bool uT,
833 const double* v, size_t vstep, bool vT,
834 const double* b, size_t bstep, int nb,
835 double* x, size_t xstep, uchar* buffer )
837 SVBkSbImpl_(m, n, w, wstep ? (int)(wstep/sizeof(w[0])) : 1,
838 u, (int)(ustep/sizeof(u[0])), uT,
839 v, (int)(vstep/sizeof(v[0])), vT,
840 b, (int)(bstep/sizeof(b[0])), nb,
841 x, (int)(xstep/sizeof(x[0])),
842 (double*)alignPtr(buffer, sizeof(double)), DBL_EPSILON*2 );
847 /****************************************************************************************\
848 * Determinant of the matrix *
849 \****************************************************************************************/
851 #define det2(m) ((double)m(0,0)*m(1,1) - (double)m(0,1)*m(1,0))
852 #define det3(m) (m(0,0)*((double)m(1,1)*m(2,2) - (double)m(1,2)*m(2,1)) - \
853 m(0,1)*((double)m(1,0)*m(2,2) - (double)m(1,2)*m(2,0)) + \
854 m(0,2)*((double)m(1,0)*m(2,1) - (double)m(1,1)*m(2,0)))
856 double cv::determinant( InputArray _mat )
858 Mat mat = _mat.getMat();
860 int type = mat.type(), rows = mat.rows;
861 size_t step = mat.step;
862 const uchar* m = mat.data;
864 CV_Assert( !mat.empty() );
865 CV_Assert( mat.rows == mat.cols && (type == CV_32F || type == CV_64F));
867 #define Mf(y, x) ((float*)(m + y*step))[x]
868 #define Md(y, x) ((double*)(m + y*step))[x]
880 size_t bufSize = rows*rows*sizeof(float);
881 AutoBuffer<uchar> buffer(bufSize);
882 Mat a(rows, rows, CV_32F, (uchar*)buffer);
885 result = LU((float*)a.data, a.step, rows, 0, 0, 0);
888 for( int i = 0; i < rows; i++ )
889 result *= ((const float*)(a.data + a.step*i))[i];
904 size_t bufSize = rows*rows*sizeof(double);
905 AutoBuffer<uchar> buffer(bufSize);
906 Mat a(rows, rows, CV_64F, (uchar*)buffer);
909 result = LU((double*)a.data, a.step, rows, 0, 0, 0);
912 for( int i = 0; i < rows; i++ )
913 result *= ((const double*)(a.data + a.step*i))[i];
925 /****************************************************************************************\
926 * Inverse (or pseudo-inverse) of a matrix *
927 \****************************************************************************************/
929 #define Sf( y, x ) ((float*)(srcdata + y*srcstep))[x]
930 #define Sd( y, x ) ((double*)(srcdata + y*srcstep))[x]
931 #define Df( y, x ) ((float*)(dstdata + y*dststep))[x]
932 #define Dd( y, x ) ((double*)(dstdata + y*dststep))[x]
934 double cv::invert( InputArray _src, OutputArray _dst, int method )
937 Mat src = _src.getMat();
938 int type = src.type();
940 CV_Assert(type == CV_32F || type == CV_64F);
942 size_t esz = CV_ELEM_SIZE(type);
943 int m = src.rows, n = src.cols;
945 if( method == DECOMP_SVD )
947 int nm = std::min(m, n);
949 AutoBuffer<uchar> _buf((m*nm + nm + nm*n)*esz + sizeof(double));
950 uchar* buf = alignPtr((uchar*)_buf, (int)esz);
951 Mat u(m, nm, type, buf);
952 Mat w(nm, 1, type, u.data + m*nm*esz);
953 Mat vt(nm, n, type, w.data + nm*esz);
955 SVD::compute(src, w, u, vt);
956 SVD::backSubst(w, u, vt, Mat(), _dst);
957 return type == CV_32F ?
958 (w.ptr<float>()[0] >= FLT_EPSILON ?
959 w.ptr<float>()[n-1]/w.ptr<float>()[0] : 0) :
960 (w.ptr<double>()[0] >= DBL_EPSILON ?
961 w.ptr<double>()[n-1]/w.ptr<double>()[0] : 0);
966 if( method == DECOMP_EIG )
968 AutoBuffer<uchar> _buf((n*n*2 + n)*esz + sizeof(double));
969 uchar* buf = alignPtr((uchar*)_buf, (int)esz);
970 Mat u(n, n, type, buf);
971 Mat w(n, 1, type, u.data + n*n*esz);
972 Mat vt(n, n, type, w.data + n*esz);
976 SVD::backSubst(w, u, vt, Mat(), _dst);
977 return type == CV_32F ?
978 (w.ptr<float>()[0] >= FLT_EPSILON ?
979 w.ptr<float>()[n-1]/w.ptr<float>()[0] : 0) :
980 (w.ptr<double>()[0] >= DBL_EPSILON ?
981 w.ptr<double>()[n-1]/w.ptr<double>()[0] : 0);
984 CV_Assert( method == DECOMP_LU || method == DECOMP_CHOLESKY );
986 _dst.create( n, n, type );
987 Mat dst = _dst.getMat();
991 const uchar* srcdata = src.data;
992 uchar* dstdata = dst.data;
993 size_t srcstep = src.step;
994 size_t dststep = dst.step;
998 if( type == CV_32FC1 )
1000 double d = det2(Sf);
1009 __m128 zero = _mm_setzero_ps();
1010 __m128 t0 = _mm_loadl_pi(zero, (const __m64*)srcdata); //t0 = sf(0,0) sf(0,1)
1011 __m128 t1 = _mm_loadh_pi(zero, (const __m64*)(srcdata+srcstep)); //t1 = sf(1,0) sf(1,1)
1012 __m128 s0 = _mm_or_ps(t0, t1);
1013 __m128 det =_mm_set1_ps((float)d);
1014 s0 = _mm_mul_ps(s0, det);
1015 static const uchar CV_DECL_ALIGNED(16) inv[16] = {0,0,0,0,0,0,0,0x80,0,0,0,0x80,0,0,0,0};
1016 __m128 pattern = _mm_load_ps((const float*)inv);
1017 s0 = _mm_xor_ps(s0, pattern);//==-1*s0
1018 s0 = _mm_shuffle_ps(s0, s0, _MM_SHUFFLE(0,2,1,3));
1019 _mm_storel_pi((__m64*)dstdata, s0);
1020 _mm_storeh_pi((__m64*)((float*)(dstdata+dststep)), s0);
1028 Df(1,1) = (float)t0;
1029 Df(0,0) = (float)t1;
1032 Df(0,1) = (float)t0;
1033 Df(1,0) = (float)t1;
1040 double d = det2(Sd);
1048 __m128d s0 = _mm_loadu_pd((const double*)srcdata); //s0 = sf(0,0) sf(0,1)
1049 __m128d s1 = _mm_loadu_pd ((const double*)(srcdata+srcstep));//s1 = sf(1,0) sf(1,1)
1050 __m128d sm = _mm_unpacklo_pd(s0, _mm_load_sd((const double*)(srcdata+srcstep)+1)); //sm = sf(0,0) sf(1,1) - main diagonal
1051 __m128d ss = _mm_shuffle_pd(s0, s1, _MM_SHUFFLE2(0,1)); //ss = sf(0,1) sf(1,0) - secondary diagonal
1052 __m128d det = _mm_load1_pd((const double*)&d);
1053 sm = _mm_mul_pd(sm, det);
1055 static const uchar CV_DECL_ALIGNED(16) inv[8] = {0,0,0,0,0,0,0,0x80};
1056 __m128d pattern = _mm_load1_pd((double*)inv);
1057 ss = _mm_mul_pd(ss, det);
1058 ss = _mm_xor_pd(ss, pattern);//==-1*ss
1060 s0 = _mm_shuffle_pd(sm, ss, _MM_SHUFFLE2(0,1));
1061 s1 = _mm_shuffle_pd(ss, sm, _MM_SHUFFLE2(0,1));
1062 _mm_storeu_pd((double*)dstdata, s0);
1063 _mm_storeu_pd((double*)(dstdata+dststep), s1);
1083 if( type == CV_32FC1 )
1085 double d = det3(Sf);
1093 t[0] = (((double)Sf(1,1) * Sf(2,2) - (double)Sf(1,2) * Sf(2,1)) * d);
1094 t[1] = (((double)Sf(0,2) * Sf(2,1) - (double)Sf(0,1) * Sf(2,2)) * d);
1095 t[2] = (((double)Sf(0,1) * Sf(1,2) - (double)Sf(0,2) * Sf(1,1)) * d);
1097 t[3] = (((double)Sf(1,2) * Sf(2,0) - (double)Sf(1,0) * Sf(2,2)) * d);
1098 t[4] = (((double)Sf(0,0) * Sf(2,2) - (double)Sf(0,2) * Sf(2,0)) * d);
1099 t[5] = (((double)Sf(0,2) * Sf(1,0) - (double)Sf(0,0) * Sf(1,2)) * d);
1101 t[6] = (((double)Sf(1,0) * Sf(2,1) - (double)Sf(1,1) * Sf(2,0)) * d);
1102 t[7] = (((double)Sf(0,1) * Sf(2,0) - (double)Sf(0,0) * Sf(2,1)) * d);
1103 t[8] = (((double)Sf(0,0) * Sf(1,1) - (double)Sf(0,1) * Sf(1,0)) * d);
1105 Df(0,0) = (float)t[0]; Df(0,1) = (float)t[1]; Df(0,2) = (float)t[2];
1106 Df(1,0) = (float)t[3]; Df(1,1) = (float)t[4]; Df(1,2) = (float)t[5];
1107 Df(2,0) = (float)t[6]; Df(2,1) = (float)t[7]; Df(2,2) = (float)t[8];
1112 double d = det3(Sd);
1119 t[0] = (Sd(1,1) * Sd(2,2) - Sd(1,2) * Sd(2,1)) * d;
1120 t[1] = (Sd(0,2) * Sd(2,1) - Sd(0,1) * Sd(2,2)) * d;
1121 t[2] = (Sd(0,1) * Sd(1,2) - Sd(0,2) * Sd(1,1)) * d;
1123 t[3] = (Sd(1,2) * Sd(2,0) - Sd(1,0) * Sd(2,2)) * d;
1124 t[4] = (Sd(0,0) * Sd(2,2) - Sd(0,2) * Sd(2,0)) * d;
1125 t[5] = (Sd(0,2) * Sd(1,0) - Sd(0,0) * Sd(1,2)) * d;
1127 t[6] = (Sd(1,0) * Sd(2,1) - Sd(1,1) * Sd(2,0)) * d;
1128 t[7] = (Sd(0,1) * Sd(2,0) - Sd(0,0) * Sd(2,1)) * d;
1129 t[8] = (Sd(0,0) * Sd(1,1) - Sd(0,1) * Sd(1,0)) * d;
1131 Dd(0,0) = t[0]; Dd(0,1) = t[1]; Dd(0,2) = t[2];
1132 Dd(1,0) = t[3]; Dd(1,1) = t[4]; Dd(1,2) = t[5];
1133 Dd(2,0) = t[6]; Dd(2,1) = t[7]; Dd(2,2) = t[8];
1141 if( type == CV_32FC1 )
1147 Df(0,0) = (float)(1./d);
1165 int elem_size = CV_ELEM_SIZE(type);
1166 AutoBuffer<uchar> buf(n*n*elem_size);
1167 Mat src1(n, n, type, (uchar*)buf);
1171 if( method == DECOMP_LU && type == CV_32F )
1172 result = LU((float*)src1.data, src1.step, n, (float*)dst.data, dst.step, n) != 0;
1173 else if( method == DECOMP_LU && type == CV_64F )
1174 result = LU((double*)src1.data, src1.step, n, (double*)dst.data, dst.step, n) != 0;
1175 else if( method == DECOMP_CHOLESKY && type == CV_32F )
1176 result = Cholesky((float*)src1.data, src1.step, n, (float*)dst.data, dst.step, n);
1178 result = Cholesky((double*)src1.data, src1.step, n, (double*)dst.data, dst.step, n);
1188 /****************************************************************************************\
1189 * Solving a linear system *
1190 \****************************************************************************************/
1192 bool cv::solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int method )
1195 Mat src = _src.getMat(), _src2 = _src2arg.getMat();
1196 int type = src.type();
1197 bool is_normal = (method & DECOMP_NORMAL) != 0;
1199 CV_Assert( type == _src2.type() && (type == CV_32F || type == CV_64F) );
1201 method &= ~DECOMP_NORMAL;
1202 CV_Assert( (method != DECOMP_LU && method != DECOMP_CHOLESKY) ||
1203 is_normal || src.rows == src.cols );
1205 // check case of a single equation and small matrix
1206 if( (method == DECOMP_LU || method == DECOMP_CHOLESKY) && !is_normal &&
1207 src.rows <= 3 && src.rows == src.cols && _src2.cols == 1 )
1209 _dst.create( src.cols, _src2.cols, src.type() );
1210 Mat dst = _dst.getMat();
1212 #define bf(y) ((float*)(bdata + y*src2step))[0]
1213 #define bd(y) ((double*)(bdata + y*src2step))[0]
1215 const uchar* srcdata = src.data;
1216 const uchar* bdata = _src2.data;
1217 uchar* dstdata = dst.data;
1218 size_t srcstep = src.step;
1219 size_t src2step = _src2.step;
1220 size_t dststep = dst.step;
1224 if( type == CV_32FC1 )
1226 double d = det2(Sf);
1231 t = (float)(((double)bf(0)*Sf(1,1) - (double)bf(1)*Sf(0,1))*d);
1232 Df(1,0) = (float)(((double)bf(1)*Sf(0,0) - (double)bf(0)*Sf(1,0))*d);
1240 double d = det2(Sd);
1245 t = (bd(0)*Sd(1,1) - bd(1)*Sd(0,1))*d;
1246 Dd(1,0) = (bd(1)*Sd(0,0) - bd(0)*Sd(1,0))*d;
1253 else if( src.rows == 3 )
1255 if( type == CV_32FC1 )
1257 double d = det3(Sf);
1264 (bf(0)*((double)Sf(1,1)*Sf(2,2) - (double)Sf(1,2)*Sf(2,1)) -
1265 Sf(0,1)*((double)bf(1)*Sf(2,2) - (double)Sf(1,2)*bf(2)) +
1266 Sf(0,2)*((double)bf(1)*Sf(2,1) - (double)Sf(1,1)*bf(2))));
1269 (Sf(0,0)*(double)(bf(1)*Sf(2,2) - (double)Sf(1,2)*bf(2)) -
1270 bf(0)*((double)Sf(1,0)*Sf(2,2) - (double)Sf(1,2)*Sf(2,0)) +
1271 Sf(0,2)*((double)Sf(1,0)*bf(2) - (double)bf(1)*Sf(2,0))));
1274 (Sf(0,0)*((double)Sf(1,1)*bf(2) - (double)bf(1)*Sf(2,1)) -
1275 Sf(0,1)*((double)Sf(1,0)*bf(2) - (double)bf(1)*Sf(2,0)) +
1276 bf(0)*((double)Sf(1,0)*Sf(2,1) - (double)Sf(1,1)*Sf(2,0))));
1287 double d = det3(Sd);
1294 t[0] = ((Sd(1,1) * Sd(2,2) - Sd(1,2) * Sd(2,1))*bd(0) +
1295 (Sd(0,2) * Sd(2,1) - Sd(0,1) * Sd(2,2))*bd(1) +
1296 (Sd(0,1) * Sd(1,2) - Sd(0,2) * Sd(1,1))*bd(2))*d;
1298 t[1] = ((Sd(1,2) * Sd(2,0) - Sd(1,0) * Sd(2,2))*bd(0) +
1299 (Sd(0,0) * Sd(2,2) - Sd(0,2) * Sd(2,0))*bd(1) +
1300 (Sd(0,2) * Sd(1,0) - Sd(0,0) * Sd(1,2))*bd(2))*d;
1302 t[2] = ((Sd(1,0) * Sd(2,1) - Sd(1,1) * Sd(2,0))*bd(0) +
1303 (Sd(0,1) * Sd(2,0) - Sd(0,0) * Sd(2,1))*bd(1) +
1304 (Sd(0,0) * Sd(1,1) - Sd(0,1) * Sd(1,0))*bd(2))*d;
1316 assert( src.rows == 1 );
1318 if( type == CV_32FC1 )
1322 Df(0,0) = (float)(bf(0)/d);
1330 Dd(0,0) = (bd(0)/d);
1338 if( method == DECOMP_QR )
1339 method = DECOMP_SVD;
1341 int m = src.rows, m_ = m, n = src.cols, nb = _src2.cols;
1342 size_t esz = CV_ELEM_SIZE(type), bufsize = 0;
1343 size_t vstep = alignSize(n*esz, 16);
1344 size_t astep = method == DECOMP_SVD && !is_normal ? alignSize(m*esz, 16) : vstep;
1345 AutoBuffer<uchar> buffer;
1348 _dst.create( src.cols, src2.cols, src.type() );
1349 Mat dst = _dst.getMat();
1352 CV_Error(CV_StsBadArg, "The function can not solve under-determined linear systems" );
1356 else if( is_normal )
1359 if( method == DECOMP_SVD )
1360 method = DECOMP_EIG;
1363 size_t asize = astep*(method == DECOMP_SVD || is_normal ? n : m);
1364 bufsize += asize + 32;
1367 bufsize += n*nb*esz;
1369 if( method == DECOMP_SVD || method == DECOMP_EIG )
1370 bufsize += n*5*esz + n*vstep + nb*sizeof(double) + 32;
1372 buffer.allocate(bufsize);
1373 uchar* ptr = alignPtr((uchar*)buffer, 16);
1375 Mat a(m_, n, type, ptr, astep);
1378 mulTransposed(src, a, true);
1379 else if( method != DECOMP_SVD )
1383 a = Mat(n, m_, type, ptr, astep);
1390 if( method == DECOMP_LU || method == DECOMP_CHOLESKY )
1396 if( method == DECOMP_LU || method == DECOMP_CHOLESKY )
1397 gemm( src, src2, 1, Mat(), 0, dst, GEMM_1_T );
1400 Mat tmp(n, nb, type, ptr);
1402 gemm( src, src2, 1, Mat(), 0, tmp, GEMM_1_T );
1407 if( method == DECOMP_LU )
1409 if( type == CV_32F )
1410 result = LU(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb) != 0;
1412 result = LU(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb) != 0;
1414 else if( method == DECOMP_CHOLESKY )
1416 if( type == CV_32F )
1417 result = Cholesky(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb);
1419 result = Cholesky(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb);
1423 ptr = alignPtr(ptr, 16);
1424 Mat v(n, n, type, ptr, vstep), w(n, 1, type, ptr + vstep*n), u;
1425 ptr += n*(vstep + esz);
1427 if( method == DECOMP_EIG )
1429 if( type == CV_32F )
1430 Jacobi(a.ptr<float>(), a.step, w.ptr<float>(), v.ptr<float>(), v.step, n, ptr);
1432 Jacobi(a.ptr<double>(), a.step, w.ptr<double>(), v.ptr<double>(), v.step, n, ptr);
1437 if( type == CV_32F )
1438 JacobiSVD(a.ptr<float>(), a.step, w.ptr<float>(), v.ptr<float>(), v.step, m_, n);
1440 JacobiSVD(a.ptr<double>(), a.step, w.ptr<double>(), v.ptr<double>(), v.step, m_, n);
1444 if( type == CV_32F )
1446 SVBkSb(m_, n, w.ptr<float>(), 0, u.ptr<float>(), u.step, true,
1447 v.ptr<float>(), v.step, true, src2.ptr<float>(),
1448 src2.step, nb, dst.ptr<float>(), dst.step, ptr);
1452 SVBkSb(m_, n, w.ptr<double>(), 0, u.ptr<double>(), u.step, true,
1453 v.ptr<double>(), v.step, true, src2.ptr<double>(),
1454 src2.step, nb, dst.ptr<double>(), dst.step, ptr);
1466 /////////////////// finding eigenvalues and eigenvectors of a symmetric matrix ///////////////
1468 bool cv::eigen( InputArray _src, OutputArray _evals, OutputArray _evects )
1470 Mat src = _src.getMat();
1471 int type = src.type();
1474 CV_Assert( src.rows == src.cols );
1475 CV_Assert (type == CV_32F || type == CV_64F);
1478 if( _evects.needed() )
1480 _evects.create(n, n, type);
1481 v = _evects.getMat();
1484 size_t elemSize = src.elemSize(), astep = alignSize(n*elemSize, 16);
1485 AutoBuffer<uchar> buf(n*astep + n*5*elemSize + 32);
1486 uchar* ptr = alignPtr((uchar*)buf, 16);
1487 Mat a(n, n, type, ptr, astep), w(n, 1, type, ptr + astep*n);
1488 ptr += astep*n + elemSize*n;
1490 bool ok = type == CV_32F ?
1491 Jacobi(a.ptr<float>(), a.step, w.ptr<float>(), v.ptr<float>(), v.step, n, ptr) :
1492 Jacobi(a.ptr<double>(), a.step, w.ptr<double>(), v.ptr<double>(), v.step, n, ptr);
1501 static void _SVDcompute( InputArray _aarr, OutputArray _w,
1502 OutputArray _u, OutputArray _vt, int flags )
1504 Mat src = _aarr.getMat();
1505 int m = src.rows, n = src.cols;
1506 int type = src.type();
1507 bool compute_uv = _u.needed() || _vt.needed();
1508 bool full_uv = (flags & SVD::FULL_UV) != 0;
1510 CV_Assert( type == CV_32F || type == CV_64F );
1512 if( flags & SVD::NO_UV )
1516 compute_uv = full_uv = false;
1526 int urows = full_uv ? m : n;
1527 size_t esz = src.elemSize(), astep = alignSize(m*esz, 16), vstep = alignSize(n*esz, 16);
1528 AutoBuffer<uchar> _buf(urows*astep + n*vstep + n*esz + 32);
1529 uchar* buf = alignPtr((uchar*)_buf, 16);
1530 Mat temp_a(n, m, type, buf, astep);
1531 Mat temp_w(n, 1, type, buf + urows*astep);
1532 Mat temp_u(urows, m, type, buf, astep), temp_v;
1535 temp_v = Mat(n, n, type, alignPtr(buf + urows*astep + n*esz, 16), vstep);
1538 temp_u = Scalar::all(0);
1541 transpose(src, temp_a);
1545 if( type == CV_32F )
1547 JacobiSVD(temp_a.ptr<float>(), temp_u.step, temp_w.ptr<float>(),
1548 temp_v.ptr<float>(), temp_v.step, m, n, compute_uv ? urows : 0);
1552 JacobiSVD(temp_a.ptr<double>(), temp_u.step, temp_w.ptr<double>(),
1553 temp_v.ptr<double>(), temp_v.step, m, n, compute_uv ? urows : 0);
1561 transpose(temp_u, _u);
1568 transpose(temp_v, _u);
1576 void SVD::compute( InputArray a, OutputArray w, OutputArray u, OutputArray vt, int flags )
1578 _SVDcompute(a, w, u, vt, flags);
1581 void SVD::compute( InputArray a, OutputArray w, int flags )
1583 _SVDcompute(a, w, noArray(), noArray(), flags);
1586 void SVD::backSubst( InputArray _w, InputArray _u, InputArray _vt,
1587 InputArray _rhs, OutputArray _dst )
1589 Mat w = _w.getMat(), u = _u.getMat(), vt = _vt.getMat(), rhs = _rhs.getMat();
1590 int type = w.type(), esz = (int)w.elemSize();
1591 int m = u.rows, n = vt.cols, nb = rhs.data ? rhs.cols : m, nm = std::min(m, n);
1592 size_t wstep = w.rows == 1 ? (size_t)esz : w.cols == 1 ? (size_t)w.step : (size_t)w.step + esz;
1593 AutoBuffer<uchar> buffer(nb*sizeof(double) + 16);
1594 CV_Assert( w.type() == u.type() && u.type() == vt.type() && u.data && vt.data && w.data );
1595 CV_Assert( u.cols >= nm && vt.rows >= nm &&
1596 (w.size() == Size(nm, 1) || w.size() == Size(1, nm) || w.size() == Size(vt.rows, u.cols)) );
1597 CV_Assert( rhs.data == 0 || (rhs.type() == type && rhs.rows == m) );
1599 _dst.create( n, nb, type );
1600 Mat dst = _dst.getMat();
1601 if( type == CV_32F )
1602 SVBkSb(m, n, w.ptr<float>(), wstep, u.ptr<float>(), u.step, false,
1603 vt.ptr<float>(), vt.step, true, rhs.ptr<float>(), rhs.step, nb,
1604 dst.ptr<float>(), dst.step, buffer);
1605 else if( type == CV_64F )
1606 SVBkSb(m, n, w.ptr<double>(), wstep, u.ptr<double>(), u.step, false,
1607 vt.ptr<double>(), vt.step, true, rhs.ptr<double>(), rhs.step, nb,
1608 dst.ptr<double>(), dst.step, buffer);
1610 CV_Error( CV_StsUnsupportedFormat, "" );
1614 SVD& SVD::operator ()(InputArray a, int flags)
1616 _SVDcompute(a, w, u, vt, flags);
1621 void SVD::backSubst( InputArray rhs, OutputArray dst ) const
1623 backSubst( w, u, vt, rhs, dst );
1629 void cv::SVDecomp(InputArray src, OutputArray w, OutputArray u, OutputArray vt, int flags)
1631 SVD::compute(src, w, u, vt, flags);
1634 void cv::SVBackSubst(InputArray w, InputArray u, InputArray vt, InputArray rhs, OutputArray dst)
1636 SVD::backSubst(w, u, vt, rhs, dst);
1641 cvDet( const CvArr* arr )
1643 if( CV_IS_MAT(arr) && ((CvMat*)arr)->rows <= 3 )
1645 CvMat* mat = (CvMat*)arr;
1646 int type = CV_MAT_TYPE(mat->type);
1647 int rows = mat->rows;
1648 uchar* m = mat->data.ptr;
1649 int step = mat->step;
1650 CV_Assert( rows == mat->cols );
1652 #define Mf(y, x) ((float*)(m + y*step))[x]
1653 #define Md(y, x) ((double*)(m + y*step))[x]
1655 if( type == CV_32F )
1662 else if( type == CV_64F )
1670 return cv::determinant(cv::cvarrToMat(arr));
1675 cvInvert( const CvArr* srcarr, CvArr* dstarr, int method )
1677 cv::Mat src = cv::cvarrToMat(srcarr), dst = cv::cvarrToMat(dstarr);
1679 CV_Assert( src.type() == dst.type() && src.rows == dst.cols && src.cols == dst.rows );
1680 return cv::invert( src, dst, method == CV_CHOLESKY ? cv::DECOMP_CHOLESKY :
1681 method == CV_SVD ? cv::DECOMP_SVD :
1682 method == CV_SVD_SYM ? cv::DECOMP_EIG : cv::DECOMP_LU );
1687 cvSolve( const CvArr* Aarr, const CvArr* barr, CvArr* xarr, int method )
1689 cv::Mat A = cv::cvarrToMat(Aarr), b = cv::cvarrToMat(barr), x = cv::cvarrToMat(xarr);
1691 CV_Assert( A.type() == x.type() && A.cols == x.rows && x.cols == b.cols );
1692 bool is_normal = (method & CV_NORMAL) != 0;
1693 method &= ~CV_NORMAL;
1694 return cv::solve( A, b, x, (method == CV_CHOLESKY ? cv::DECOMP_CHOLESKY :
1695 method == CV_SVD ? cv::DECOMP_SVD :
1696 method == CV_SVD_SYM ? cv::DECOMP_EIG :
1697 A.rows > A.cols ? cv::DECOMP_QR : cv::DECOMP_LU) + (is_normal ? cv::DECOMP_NORMAL : 0) );
1702 cvEigenVV( CvArr* srcarr, CvArr* evectsarr, CvArr* evalsarr, double,
1705 cv::Mat src = cv::cvarrToMat(srcarr), evals0 = cv::cvarrToMat(evalsarr), evals = evals0;
1708 cv::Mat evects0 = cv::cvarrToMat(evectsarr), evects = evects0;
1709 eigen(src, evals, evects);
1710 if( evects0.data != evects.data )
1712 const uchar* p = evects0.data;
1713 evects.convertTo(evects0, evects0.type());
1714 CV_Assert( p == evects0.data );
1719 if( evals0.data != evals.data )
1721 const uchar* p = evals0.data;
1722 if( evals0.size() == evals.size() )
1723 evals.convertTo(evals0, evals0.type());
1724 else if( evals0.type() == evals.type() )
1725 cv::transpose(evals, evals0);
1727 cv::Mat(evals.t()).convertTo(evals0, evals0.type());
1728 CV_Assert( p == evals0.data );
1734 cvSVD( CvArr* aarr, CvArr* warr, CvArr* uarr, CvArr* varr, int flags )
1736 cv::Mat a = cv::cvarrToMat(aarr), w = cv::cvarrToMat(warr), u, v;
1737 int m = a.rows, n = a.cols, type = a.type(), mn = std::max(m, n), nm = std::min(m, n);
1739 CV_Assert( w.type() == type &&
1740 (w.size() == cv::Size(nm,1) || w.size() == cv::Size(1, nm) ||
1741 w.size() == cv::Size(nm, nm) || w.size() == cv::Size(n, m)) );
1745 if( w.size() == cv::Size(nm, 1) )
1746 svd.w = cv::Mat(nm, 1, type, w.data );
1747 else if( w.isContinuous() )
1752 u = cv::cvarrToMat(uarr);
1753 CV_Assert( u.type() == type );
1759 v = cv::cvarrToMat(varr);
1760 CV_Assert( v.type() == type );
1764 svd(a, ((flags & CV_SVD_MODIFY_A) ? cv::SVD::MODIFY_A : 0) |
1765 ((!svd.u.data && !svd.vt.data) ? cv::SVD::NO_UV : 0) |
1766 ((m != n && (svd.u.size() == cv::Size(mn, mn) ||
1767 svd.vt.size() == cv::Size(mn, mn))) ? cv::SVD::FULL_UV : 0));
1771 if( flags & CV_SVD_U_T )
1772 cv::transpose( svd.u, u );
1773 else if( u.data != svd.u.data )
1775 CV_Assert( u.size() == svd.u.size() );
1782 if( !(flags & CV_SVD_V_T) )
1783 cv::transpose( svd.vt, v );
1784 else if( v.data != svd.vt.data )
1786 CV_Assert( v.size() == svd.vt.size() );
1791 if( w.data != svd.w.data )
1793 if( w.size() == svd.w.size() )
1798 cv::Mat wd = w.diag();
1806 cvSVBkSb( const CvArr* warr, const CvArr* uarr,
1807 const CvArr* varr, const CvArr* rhsarr,
1808 CvArr* dstarr, int flags )
1810 cv::Mat w = cv::cvarrToMat(warr), u = cv::cvarrToMat(uarr),
1811 v = cv::cvarrToMat(varr), rhs,
1812 dst = cv::cvarrToMat(dstarr), dst0 = dst;
1813 if( flags & CV_SVD_U_T )
1819 if( !(flags & CV_SVD_V_T) )
1826 rhs = cv::cvarrToMat(rhsarr);
1828 cv::SVD::backSubst(w, u, v, rhs, dst);
1829 CV_Assert( dst.data == dst0.data );