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 #ifndef __OPENCV_FEATURES_2D_HPP__
44 #define __OPENCV_FEATURES_2D_HPP__
46 #include "opencv2/core.hpp"
47 #include "opencv2/flann/miniflann.hpp"
52 CV_EXPORTS bool initModule_features2d();
54 // //! writes vector of keypoints to the file storage
55 // CV_EXPORTS void write(FileStorage& fs, const String& name, const std::vector<KeyPoint>& keypoints);
56 // //! reads vector of keypoints from the specified file storage node
57 // CV_EXPORTS void read(const FileNode& node, CV_OUT std::vector<KeyPoint>& keypoints);
60 * A class filters a vector of keypoints.
61 * Because now it is difficult to provide a convenient interface for all usage scenarios of the keypoints filter class,
62 * it has only several needed by now static methods.
64 class CV_EXPORTS KeyPointsFilter
70 * Remove keypoints within borderPixels of an image edge.
72 static void runByImageBorder( std::vector<KeyPoint>& keypoints, Size imageSize, int borderSize );
74 * Remove keypoints of sizes out of range.
76 static void runByKeypointSize( std::vector<KeyPoint>& keypoints, float minSize,
77 float maxSize=FLT_MAX );
79 * Remove keypoints from some image by mask for pixels of this image.
81 static void runByPixelsMask( std::vector<KeyPoint>& keypoints, const Mat& mask );
83 * Remove duplicated keypoints.
85 static void removeDuplicated( std::vector<KeyPoint>& keypoints );
88 * Retain the specified number of the best keypoints (according to the response)
90 static void retainBest( std::vector<KeyPoint>& keypoints, int npoints );
94 /************************************ Base Classes ************************************/
97 * Abstract base class for 2D image feature detectors.
99 class CV_EXPORTS_W FeatureDetector : public virtual Algorithm
102 virtual ~FeatureDetector();
105 * Detect keypoints in an image.
107 * keypoints The detected keypoints.
108 * mask Mask specifying where to look for keypoints (optional). Must be a char
109 * matrix with non-zero values in the region of interest.
111 CV_WRAP void detect( InputArray image, CV_OUT std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
114 * Detect keypoints in an image set.
115 * images Image collection.
116 * keypoints Collection of keypoints detected in an input images. keypoints[i] is a set of keypoints detected in an images[i].
117 * masks Masks for image set. masks[i] is a mask for images[i].
119 void detect( InputArrayOfArrays images, std::vector<std::vector<KeyPoint> >& keypoints, InputArrayOfArrays masks=noArray() ) const;
121 // Return true if detector object is empty
122 CV_WRAP virtual bool empty() const;
124 // Create feature detector by detector name.
125 CV_WRAP static Ptr<FeatureDetector> create( const String& detectorType );
128 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const = 0;
131 * Remove keypoints that are not in the mask.
132 * Helper function, useful when wrapping a library call for keypoint detection that
133 * does not support a mask argument.
135 static void removeInvalidPoints( const Mat & mask, std::vector<KeyPoint>& keypoints );
140 * Abstract base class for computing descriptors for image keypoints.
142 * In this interface we assume a keypoint descriptor can be represented as a
143 * dense, fixed-dimensional vector of some basic type. Most descriptors used
144 * in practice follow this pattern, as it makes it very easy to compute
145 * distances between descriptors. Therefore we represent a collection of
146 * descriptors as a Mat, where each row is one keypoint descriptor.
148 class CV_EXPORTS_W DescriptorExtractor : public virtual Algorithm
151 virtual ~DescriptorExtractor();
154 * Compute the descriptors for a set of keypoints in an image.
156 * keypoints The input keypoints. Keypoints for which a descriptor cannot be computed are removed.
157 * descriptors Copmputed descriptors. Row i is the descriptor for keypoint i.
159 CV_WRAP void compute( InputArray image, CV_OUT CV_IN_OUT std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
162 * Compute the descriptors for a keypoints collection detected in image collection.
163 * images Image collection.
164 * keypoints Input keypoints collection. keypoints[i] is keypoints detected in images[i].
165 * Keypoints for which a descriptor cannot be computed are removed.
166 * descriptors Descriptor collection. descriptors[i] are descriptors computed for set keypoints[i].
168 void compute( InputArrayOfArrays images, std::vector<std::vector<KeyPoint> >& keypoints, OutputArrayOfArrays descriptors ) const;
170 CV_WRAP virtual int descriptorSize() const = 0;
171 CV_WRAP virtual int descriptorType() const = 0;
172 CV_WRAP virtual int defaultNorm() const = 0;
174 CV_WRAP virtual bool empty() const;
176 CV_WRAP static Ptr<DescriptorExtractor> create( const String& descriptorExtractorType );
179 virtual void computeImpl( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const = 0;
182 * Remove keypoints within borderPixels of an image edge.
184 static void removeBorderKeypoints( std::vector<KeyPoint>& keypoints,
185 Size imageSize, int borderSize );
191 * Abstract base class for simultaneous 2D feature detection descriptor extraction.
193 class CV_EXPORTS_W Feature2D : public FeatureDetector, public DescriptorExtractor
197 * Detect keypoints in an image.
199 * keypoints The detected keypoints.
200 * mask Mask specifying where to look for keypoints (optional). Must be a char
201 * matrix with non-zero values in the region of interest.
202 * useProvidedKeypoints If true, the method will skip the detection phase and will compute
203 * descriptors for the provided keypoints
205 CV_WRAP_AS(detectAndCompute) virtual void operator()( InputArray image, InputArray mask,
206 CV_OUT std::vector<KeyPoint>& keypoints,
207 OutputArray descriptors,
208 bool useProvidedKeypoints=false ) const = 0;
210 CV_WRAP void compute( InputArray image, CV_OUT CV_IN_OUT std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
212 // Create feature detector and descriptor extractor by name.
213 CV_WRAP static Ptr<Feature2D> create( const String& name );
219 class CV_EXPORTS_W BRISK : public Feature2D
222 CV_WRAP explicit BRISK(int thresh=30, int octaves=3, float patternScale=1.0f);
226 // returns the descriptor size in bytes
227 int descriptorSize() const;
228 // returns the descriptor type
229 int descriptorType() const;
230 // returns the default norm type
231 int defaultNorm() const;
233 // Compute the BRISK features on an image
234 void operator()(InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints) const;
236 // Compute the BRISK features and descriptors on an image
237 void operator()( InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints,
238 OutputArray descriptors, bool useProvidedKeypoints=false ) const;
240 AlgorithmInfo* info() const;
243 CV_WRAP explicit BRISK(std::vector<float> &radiusList, std::vector<int> &numberList,
244 float dMax=5.85f, float dMin=8.2f, std::vector<int> indexChange=std::vector<int>());
246 // call this to generate the kernel:
247 // circle of radius r (pixels), with n points;
248 // short pairings with dMax, long pairings with dMin
249 CV_WRAP void generateKernel(std::vector<float> &radiusList,
250 std::vector<int> &numberList, float dMax=5.85f, float dMin=8.2f,
251 std::vector<int> indexChange=std::vector<int>());
255 void computeImpl( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
256 void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
258 void computeKeypointsNoOrientation(InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints) const;
259 void computeDescriptorsAndOrOrientation(InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints,
260 OutputArray descriptors, bool doDescriptors, bool doOrientation,
261 bool useProvidedKeypoints) const;
263 // Feature parameters
264 CV_PROP_RW int threshold;
265 CV_PROP_RW int octaves;
267 // some helper structures for the Brisk pattern representation
268 struct BriskPatternPoint{
269 float x; // x coordinate relative to center
270 float y; // x coordinate relative to center
271 float sigma; // Gaussian smoothing sigma
273 struct BriskShortPair{
274 unsigned int i; // index of the first pattern point
275 unsigned int j; // index of other pattern point
277 struct BriskLongPair{
278 unsigned int i; // index of the first pattern point
279 unsigned int j; // index of other pattern point
280 int weighted_dx; // 1024.0/dx
281 int weighted_dy; // 1024.0/dy
283 inline int smoothedIntensity(const cv::Mat& image,
284 const cv::Mat& integral,const float key_x,
285 const float key_y, const unsigned int scale,
286 const unsigned int rot, const unsigned int point) const;
287 // pattern properties
288 BriskPatternPoint* patternPoints_; //[i][rotation][scale]
289 unsigned int points_; // total number of collocation points
290 float* scaleList_; // lists the scaling per scale index [scale]
291 unsigned int* sizeList_; // lists the total pattern size per scale index [scale]
292 static const unsigned int scales_; // scales discretization
293 static const float scalerange_; // span of sizes 40->4 Octaves - else, this needs to be adjusted...
294 static const unsigned int n_rot_; // discretization of the rotation look-up
297 int strings_; // number of uchars the descriptor consists of
298 float dMax_; // short pair maximum distance
299 float dMin_; // long pair maximum distance
300 BriskShortPair* shortPairs_; // d<_dMax
301 BriskLongPair* longPairs_; // d>_dMin
302 unsigned int noShortPairs_; // number of shortParis
303 unsigned int noLongPairs_; // number of longParis
306 static const float basicSize_;
313 class CV_EXPORTS_W ORB : public Feature2D
316 // the size of the signature in bytes
317 enum { kBytes = 32, HARRIS_SCORE=0, FAST_SCORE=1 };
319 CV_WRAP explicit ORB(int nfeatures = 500, float scaleFactor = 1.2f, int nlevels = 8, int edgeThreshold = 31,
320 int firstLevel = 0, int WTA_K=2, int scoreType=ORB::HARRIS_SCORE, int patchSize=31 );
322 // returns the descriptor size in bytes
323 int descriptorSize() const;
324 // returns the descriptor type
325 int descriptorType() const;
326 // returns the default norm type
327 int defaultNorm() const;
329 // Compute the ORB features and descriptors on an image
330 void operator()(InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints) const;
332 // Compute the ORB features and descriptors on an image
333 void operator()( InputArray image, InputArray mask, std::vector<KeyPoint>& keypoints,
334 OutputArray descriptors, bool useProvidedKeypoints=false ) const;
336 AlgorithmInfo* info() const;
340 void computeImpl( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
341 void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
343 CV_PROP_RW int nfeatures;
344 CV_PROP_RW double scaleFactor;
345 CV_PROP_RW int nlevels;
346 CV_PROP_RW int edgeThreshold;
347 CV_PROP_RW int firstLevel;
348 CV_PROP_RW int WTA_K;
349 CV_PROP_RW int scoreType;
350 CV_PROP_RW int patchSize;
353 typedef ORB OrbFeatureDetector;
354 typedef ORB OrbDescriptorExtractor;
359 class CV_EXPORTS FREAK : public DescriptorExtractor
363 * @param orientationNormalized enable orientation normalization
364 * @param scaleNormalized enable scale normalization
365 * @param patternScale scaling of the description pattern
366 * @param nbOctave number of octaves covered by the detected keypoints
367 * @param selectedPairs (optional) user defined selected pairs
369 explicit FREAK( bool orientationNormalized = true,
370 bool scaleNormalized = true,
371 float patternScale = 22.0f,
373 const std::vector<int>& selectedPairs = std::vector<int>());
374 FREAK( const FREAK& rhs );
375 FREAK& operator=( const FREAK& );
379 /** returns the descriptor length in bytes */
380 virtual int descriptorSize() const;
382 /** returns the descriptor type */
383 virtual int descriptorType() const;
385 /** returns the default norm type */
386 virtual int defaultNorm() const;
388 /** select the 512 "best description pairs"
389 * @param images grayscale images set
390 * @param keypoints set of detected keypoints
391 * @param corrThresh correlation threshold
392 * @param verbose print construction information
393 * @return list of best pair indexes
395 std::vector<int> selectPairs( const std::vector<Mat>& images, std::vector<std::vector<KeyPoint> >& keypoints,
396 const double corrThresh = 0.7, bool verbose = true );
398 AlgorithmInfo* info() const;
402 NB_SCALES = 64, NB_PAIRS = 512, NB_ORIENPAIRS = 45
406 virtual void computeImpl( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
408 uchar meanIntensity( InputArray image, InputArray integral, const float kp_x, const float kp_y,
409 const unsigned int scale, const unsigned int rot, const unsigned int point ) const;
411 bool orientationNormalized; //true if the orientation is normalized, false otherwise
412 bool scaleNormalized; //true if the scale is normalized, false otherwise
413 double patternScale; //scaling of the pattern
414 int nOctaves; //number of octaves
415 bool extAll; // true if all pairs need to be extracted for pairs selection
417 double patternScale0;
419 std::vector<int> selectedPairs0;
423 float x; // x coordinate relative to center
424 float y; // x coordinate relative to center
425 float sigma; // Gaussian smoothing sigma
428 struct DescriptionPair
430 uchar i; // index of the first point
431 uchar j; // index of the second point
434 struct OrientationPair
436 uchar i; // index of the first point
437 uchar j; // index of the second point
438 int weight_dx; // dx/(norm_sq))*4096
439 int weight_dy; // dy/(norm_sq))*4096
442 std::vector<PatternPoint> patternLookup; // look-up table for the pattern points (position+sigma of all points at all scales and orientation)
443 int patternSizes[NB_SCALES]; // size of the pattern at a specific scale (used to check if a point is within image boundaries)
444 DescriptionPair descriptionPairs[NB_PAIRS];
445 OrientationPair orientationPairs[NB_ORIENPAIRS];
450 Maximal Stable Extremal Regions class.
452 The class implements MSER algorithm introduced by J. Matas.
453 Unlike SIFT, SURF and many other detectors in OpenCV, this is salient region detector,
454 not the salient point detector.
456 It returns the regions, each of those is encoded as a contour.
458 class CV_EXPORTS_W MSER : public FeatureDetector
461 //! the full constructor
462 CV_WRAP explicit MSER( int _delta=5, int _min_area=60, int _max_area=14400,
463 double _max_variation=0.25, double _min_diversity=.2,
464 int _max_evolution=200, double _area_threshold=1.01,
465 double _min_margin=0.003, int _edge_blur_size=5 );
467 //! the operator that extracts the MSERs from the image or the specific part of it
468 CV_WRAP_AS(detect) void operator()( InputArray image, CV_OUT std::vector<std::vector<Point> >& msers,
469 InputArray mask=noArray() ) const;
470 AlgorithmInfo* info() const;
473 void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
481 double areaThreshold;
486 typedef MSER MserFeatureDetector;
491 The class implements the keypoint detector introduced by K. Konolige.
493 class CV_EXPORTS_W StarDetector : public FeatureDetector
496 //! the full constructor
497 CV_WRAP StarDetector(int _maxSize=45, int _responseThreshold=30,
498 int _lineThresholdProjected=10,
499 int _lineThresholdBinarized=8,
500 int _suppressNonmaxSize=5);
502 //! finds the keypoints in the image
503 CV_WRAP_AS(detect) void operator()(const Mat& image,
504 CV_OUT std::vector<KeyPoint>& keypoints) const;
506 AlgorithmInfo* info() const;
509 void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
512 int responseThreshold;
513 int lineThresholdProjected;
514 int lineThresholdBinarized;
515 int suppressNonmaxSize;
518 //! detects corners using FAST algorithm by E. Rosten
519 CV_EXPORTS void FAST( InputArray image, CV_OUT std::vector<KeyPoint>& keypoints,
520 int threshold, bool nonmaxSupression=true );
522 CV_EXPORTS void FAST( InputArray image, CV_OUT std::vector<KeyPoint>& keypoints,
523 int threshold, bool nonmaxSupression, int type );
525 class CV_EXPORTS_W FastFeatureDetector : public FeatureDetector
530 TYPE_5_8 = 0, TYPE_7_12 = 1, TYPE_9_16 = 2
533 CV_WRAP FastFeatureDetector( int threshold=10, bool nonmaxSuppression=true);
534 CV_WRAP FastFeatureDetector( int threshold, bool nonmaxSuppression, int type);
535 AlgorithmInfo* info() const;
538 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
541 bool nonmaxSuppression;
546 class CV_EXPORTS_W GFTTDetector : public FeatureDetector
549 CV_WRAP GFTTDetector( int maxCorners=1000, double qualityLevel=0.01, double minDistance=1,
550 int blockSize=3, bool useHarrisDetector=false, double k=0.04 );
551 AlgorithmInfo* info() const;
554 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
560 bool useHarrisDetector;
564 typedef GFTTDetector GoodFeaturesToTrackDetector;
565 typedef StarDetector StarFeatureDetector;
567 class CV_EXPORTS_W SimpleBlobDetector : public FeatureDetector
570 struct CV_EXPORTS_W_SIMPLE Params
573 CV_PROP_RW float thresholdStep;
574 CV_PROP_RW float minThreshold;
575 CV_PROP_RW float maxThreshold;
576 CV_PROP_RW size_t minRepeatability;
577 CV_PROP_RW float minDistBetweenBlobs;
579 CV_PROP_RW bool filterByColor;
580 CV_PROP_RW uchar blobColor;
582 CV_PROP_RW bool filterByArea;
583 CV_PROP_RW float minArea, maxArea;
585 CV_PROP_RW bool filterByCircularity;
586 CV_PROP_RW float minCircularity, maxCircularity;
588 CV_PROP_RW bool filterByInertia;
589 CV_PROP_RW float minInertiaRatio, maxInertiaRatio;
591 CV_PROP_RW bool filterByConvexity;
592 CV_PROP_RW float minConvexity, maxConvexity;
594 void read( const FileNode& fn );
595 void write( FileStorage& fs ) const;
598 CV_WRAP SimpleBlobDetector(const SimpleBlobDetector::Params ¶meters = SimpleBlobDetector::Params());
600 virtual void read( const FileNode& fn );
601 virtual void write( FileStorage& fs ) const;
604 struct CV_EXPORTS Center
611 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
612 virtual void findBlobs(InputArray image, InputArray binaryImage, std::vector<Center> ¢ers) const;
615 AlgorithmInfo* info() const;
619 class CV_EXPORTS DenseFeatureDetector : public FeatureDetector
622 explicit DenseFeatureDetector( float initFeatureScale=1.f, int featureScaleLevels=1,
623 float featureScaleMul=0.1f,
624 int initXyStep=6, int initImgBound=0,
625 bool varyXyStepWithScale=true,
626 bool varyImgBoundWithScale=false );
627 AlgorithmInfo* info() const;
630 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
632 double initFeatureScale;
633 int featureScaleLevels;
634 double featureScaleMul;
639 bool varyXyStepWithScale;
640 bool varyImgBoundWithScale;
644 * Adapts a detector to partition the source image into a grid and detect
645 * points in each cell.
647 class CV_EXPORTS_W GridAdaptedFeatureDetector : public FeatureDetector
651 * detector Detector that will be adapted.
652 * maxTotalKeypoints Maximum count of keypoints detected on the image. Only the strongest keypoints
654 * gridRows Grid rows count.
655 * gridCols Grid column count.
657 CV_WRAP GridAdaptedFeatureDetector( const Ptr<FeatureDetector>& detector=Ptr<FeatureDetector>(),
658 int maxTotalKeypoints=1000,
659 int gridRows=4, int gridCols=4 );
661 // TODO implement read/write
662 virtual bool empty() const;
664 AlgorithmInfo* info() const;
667 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
669 Ptr<FeatureDetector> detector;
670 int maxTotalKeypoints;
676 * Adapts a detector to detect points over multiple levels of a Gaussian
677 * pyramid. Useful for detectors that are not inherently scaled.
679 class CV_EXPORTS_W PyramidAdaptedFeatureDetector : public FeatureDetector
682 // maxLevel - The 0-based index of the last pyramid layer
683 CV_WRAP PyramidAdaptedFeatureDetector( const Ptr<FeatureDetector>& detector, int maxLevel=2 );
685 // TODO implement read/write
686 virtual bool empty() const;
689 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
691 Ptr<FeatureDetector> detector;
695 /** \brief A feature detector parameter adjuster, this is used by the DynamicAdaptedFeatureDetector
696 * and is a wrapper for FeatureDetector that allow them to be adjusted after a detection
698 class CV_EXPORTS AdjusterAdapter: public FeatureDetector
701 /** pure virtual interface
703 virtual ~AdjusterAdapter() {}
704 /** too few features were detected so, adjust the detector params accordingly
705 * \param min the minimum number of desired features
706 * \param n_detected the number previously detected
708 virtual void tooFew(int min, int n_detected) = 0;
709 /** too many features were detected so, adjust the detector params accordingly
710 * \param max the maximum number of desired features
711 * \param n_detected the number previously detected
713 virtual void tooMany(int max, int n_detected) = 0;
714 /** are params maxed out or still valid?
715 * \return false if the parameters can't be adjusted any more
717 virtual bool good() const = 0;
719 virtual Ptr<AdjusterAdapter> clone() const = 0;
721 static Ptr<AdjusterAdapter> create( const String& detectorType );
723 /** \brief an adaptively adjusting detector that iteratively detects until the desired number
724 * of features are detected.
725 * Beware that this is not thread safe - as the adjustment of parameters breaks the const
726 * of the detection routine...
727 * /TODO Make this const correct and thread safe
730 //will create a detector that attempts to find 100 - 110 FAST Keypoints, and will at most run
731 //FAST feature detection 10 times until that number of keypoints are found
732 Ptr<FeatureDetector> detector(new DynamicAdaptedFeatureDetector(new FastAdjuster(20,true),100, 110, 10));
735 class CV_EXPORTS DynamicAdaptedFeatureDetector: public FeatureDetector
739 /** \param adjuster an AdjusterAdapter that will do the detection and parameter adjustment
740 * \param max_features the maximum desired number of features
741 * \param max_iters the maximum number of times to try to adjust the feature detector params
742 * for the FastAdjuster this can be high, but with Star or Surf this can get time consuming
743 * \param min_features the minimum desired features
745 DynamicAdaptedFeatureDetector( const Ptr<AdjusterAdapter>& adjuster, int min_features=400, int max_features=500, int max_iters=5 );
747 virtual bool empty() const;
750 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
753 DynamicAdaptedFeatureDetector& operator=(const DynamicAdaptedFeatureDetector&);
754 DynamicAdaptedFeatureDetector(const DynamicAdaptedFeatureDetector&);
757 int min_features_, max_features_;
758 const Ptr<AdjusterAdapter> adjuster_;
761 /**\brief an adjust for the FAST detector. This will basically decrement or increment the
764 class CV_EXPORTS FastAdjuster: public AdjusterAdapter
767 /**\param init_thresh the initial threshold to start with, default = 20
768 * \param nonmax whether to use non max or not for fast feature detection
770 FastAdjuster(int init_thresh=20, bool nonmax=true, int min_thresh=1, int max_thresh=200);
772 virtual void tooFew(int minv, int n_detected);
773 virtual void tooMany(int maxv, int n_detected);
774 virtual bool good() const;
776 virtual Ptr<AdjusterAdapter> clone() const;
779 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
783 int init_thresh_, min_thresh_, max_thresh_;
787 /** An adjuster for StarFeatureDetector, this one adjusts the responseThreshold for now
788 * TODO find a faster way to converge the parameters for Star - use CvStarDetectorParams
790 class CV_EXPORTS StarAdjuster: public AdjusterAdapter
793 StarAdjuster(double initial_thresh=30.0, double min_thresh=2., double max_thresh=200.);
795 virtual void tooFew(int minv, int n_detected);
796 virtual void tooMany(int maxv, int n_detected);
797 virtual bool good() const;
799 virtual Ptr<AdjusterAdapter> clone() const;
802 virtual void detectImpl(InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
804 double thresh_, init_thresh_, min_thresh_, max_thresh_;
807 class CV_EXPORTS SurfAdjuster: public AdjusterAdapter
810 SurfAdjuster( double initial_thresh=400.f, double min_thresh=2, double max_thresh=1000 );
812 virtual void tooFew(int minv, int n_detected);
813 virtual void tooMany(int maxv, int n_detected);
814 virtual bool good() const;
816 virtual Ptr<AdjusterAdapter> clone() const;
819 virtual void detectImpl( InputArray image, std::vector<KeyPoint>& keypoints, InputArray mask=noArray() ) const;
821 double thresh_, init_thresh_, min_thresh_, max_thresh_;
824 CV_EXPORTS Mat windowedMatchingMask( const std::vector<KeyPoint>& keypoints1, const std::vector<KeyPoint>& keypoints2,
825 float maxDeltaX, float maxDeltaY );
830 * OpponentColorDescriptorExtractor
832 * Adapts a descriptor extractor to compute descripors in Opponent Color Space
833 * (refer to van de Sande et al., CGIV 2008 "Color Descriptors for Object Category Recognition").
834 * Input RGB image is transformed in Opponent Color Space. Then unadapted descriptor extractor
835 * (set in constructor) computes descriptors on each of the three channel and concatenate
836 * them into a single color descriptor.
838 class CV_EXPORTS OpponentColorDescriptorExtractor : public DescriptorExtractor
841 OpponentColorDescriptorExtractor( const Ptr<DescriptorExtractor>& descriptorExtractor );
843 virtual void read( const FileNode& );
844 virtual void write( FileStorage& ) const;
846 virtual int descriptorSize() const;
847 virtual int descriptorType() const;
848 virtual int defaultNorm() const;
850 virtual bool empty() const;
853 virtual void computeImpl( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors ) const;
855 Ptr<DescriptorExtractor> descriptorExtractor;
861 class CV_EXPORTS BriefDescriptorExtractor : public DescriptorExtractor
864 static const int PATCH_SIZE = 48;
865 static const int KERNEL_SIZE = 9;
867 // bytes is a length of descriptor in bytes. It can be equal 16, 32 or 64 bytes.
868 BriefDescriptorExtractor( int bytes = 32 );
870 virtual void read( const FileNode& );
871 virtual void write( FileStorage& ) const;
873 virtual int descriptorSize() const;
874 virtual int descriptorType() const;
875 virtual int defaultNorm() const;
877 /// @todo read and write for brief
879 AlgorithmInfo* info() const;
882 virtual void computeImpl(InputArray image, std::vector<KeyPoint>& keypoints, OutputArray descriptors) const;
884 typedef void(*PixelTestFn)(InputArray, const std::vector<KeyPoint>&, OutputArray);
887 PixelTestFn test_fn_;
891 /****************************************************************************************\
893 \****************************************************************************************/
896 struct CV_EXPORTS Accumulator
901 template<> struct Accumulator<unsigned char> { typedef float Type; };
902 template<> struct Accumulator<unsigned short> { typedef float Type; };
903 template<> struct Accumulator<char> { typedef float Type; };
904 template<> struct Accumulator<short> { typedef float Type; };
907 * Squared Euclidean distance functor
910 struct CV_EXPORTS SL2
912 enum { normType = NORM_L2SQR };
914 typedef typename Accumulator<T>::Type ResultType;
916 ResultType operator()( const T* a, const T* b, int size ) const
918 return normL2Sqr<ValueType, ResultType>(a, b, size);
923 * Euclidean distance functor
928 enum { normType = NORM_L2 };
930 typedef typename Accumulator<T>::Type ResultType;
932 ResultType operator()( const T* a, const T* b, int size ) const
934 return (ResultType)std::sqrt((double)normL2Sqr<ValueType, ResultType>(a, b, size));
939 * Manhattan distance (city block distance) functor
944 enum { normType = NORM_L1 };
946 typedef typename Accumulator<T>::Type ResultType;
948 ResultType operator()( const T* a, const T* b, int size ) const
950 return normL1<ValueType, ResultType>(a, b, size);
955 * Hamming distance functor - counts the bit differences between two strings - useful for the Brief descriptor
956 * bit count of A exclusive XOR'ed with B
958 struct CV_EXPORTS Hamming
960 enum { normType = NORM_HAMMING };
961 typedef unsigned char ValueType;
962 typedef int ResultType;
964 /** this will count the bits in a ^ b
966 ResultType operator()( const unsigned char* a, const unsigned char* b, int size ) const
968 return normHamming(a, b, size);
972 typedef Hamming HammingLUT;
974 template<int cellsize> struct HammingMultilevel
976 enum { normType = NORM_HAMMING + (cellsize>1) };
977 typedef unsigned char ValueType;
978 typedef int ResultType;
980 ResultType operator()( const unsigned char* a, const unsigned char* b, int size ) const
982 return normHamming(a, b, size, cellsize);
986 /****************************************************************************************\
987 * DescriptorMatcher *
988 \****************************************************************************************/
990 * Abstract base class for matching two sets of descriptors.
992 class CV_EXPORTS_W DescriptorMatcher : public Algorithm
995 virtual ~DescriptorMatcher();
998 * Add descriptors to train descriptor collection.
999 * descriptors Descriptors to add. Each descriptors[i] is a descriptors set from one image.
1001 CV_WRAP virtual void add( InputArrayOfArrays descriptors );
1003 * Get train descriptors collection.
1005 CV_WRAP const std::vector<Mat>& getTrainDescriptors() const;
1007 * Clear train descriptors collection.
1009 CV_WRAP virtual void clear();
1012 * Return true if there are not train descriptors in collection.
1014 CV_WRAP virtual bool empty() const;
1016 * Return true if the matcher supports mask in match methods.
1018 CV_WRAP virtual bool isMaskSupported() const = 0;
1021 * Train matcher (e.g. train flann index).
1022 * In all methods to match the method train() is run every time before matching.
1023 * Some descriptor matchers (e.g. BruteForceMatcher) have empty implementation
1024 * of this method, other matchers really train their inner structures
1025 * (e.g. FlannBasedMatcher trains flann::Index). So nonempty implementation
1026 * of train() should check the class object state and do traing/retraining
1027 * only if the state requires that (e.g. FlannBasedMatcher trains flann::Index
1028 * if it has not trained yet or if new descriptors have been added to the train
1031 CV_WRAP virtual void train();
1033 * Group of methods to match descriptors from image pair.
1034 * Method train() is run in this methods.
1036 // Find one best match for each query descriptor (if mask is empty).
1037 CV_WRAP void match( InputArray queryDescriptors, InputArray trainDescriptors,
1038 CV_OUT std::vector<DMatch>& matches, InputArray mask=noArray() ) const;
1039 // Find k best matches for each query descriptor (in increasing order of distances).
1040 // compactResult is used when mask is not empty. If compactResult is false matches
1041 // vector will have the same size as queryDescriptors rows. If compactResult is true
1042 // matches vector will not contain matches for fully masked out query descriptors.
1043 CV_WRAP void knnMatch( InputArray queryDescriptors, InputArray trainDescriptors,
1044 CV_OUT std::vector<std::vector<DMatch> >& matches, int k,
1045 InputArray mask=noArray(), bool compactResult=false ) const;
1046 // Find best matches for each query descriptor which have distance less than
1047 // maxDistance (in increasing order of distances).
1048 void radiusMatch( InputArray queryDescriptors, InputArray trainDescriptors,
1049 std::vector<std::vector<DMatch> >& matches, float maxDistance,
1050 InputArray mask=noArray(), bool compactResult=false ) const;
1052 * Group of methods to match descriptors from one image to image set.
1053 * See description of similar methods for matching image pair above.
1055 CV_WRAP void match( InputArray queryDescriptors, CV_OUT std::vector<DMatch>& matches,
1056 InputArrayOfArrays masks=noArray() );
1057 CV_WRAP void knnMatch( InputArray queryDescriptors, CV_OUT std::vector<std::vector<DMatch> >& matches, int k,
1058 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1059 void radiusMatch( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, float maxDistance,
1060 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1062 // Reads matcher object from a file node
1063 virtual void read( const FileNode& );
1064 // Writes matcher object to a file storage
1065 virtual void write( FileStorage& ) const;
1067 // Clone the matcher. If emptyTrainData is false the method create deep copy of the object, i.e. copies
1068 // both parameters and train data. If emptyTrainData is true the method create object copy with current parameters
1069 // but with empty train data.
1070 virtual Ptr<DescriptorMatcher> clone( bool emptyTrainData=false ) const = 0;
1072 CV_WRAP static Ptr<DescriptorMatcher> create( const String& descriptorMatcherType );
1075 * Class to work with descriptors from several images as with one merged matrix.
1076 * It is used e.g. in FlannBasedMatcher.
1078 class CV_EXPORTS DescriptorCollection
1081 DescriptorCollection();
1082 DescriptorCollection( const DescriptorCollection& collection );
1083 virtual ~DescriptorCollection();
1085 // Vector of matrices "descriptors" will be merged to one matrix "mergedDescriptors" here.
1086 void set( const std::vector<Mat>& descriptors );
1087 virtual void clear();
1089 const Mat& getDescriptors() const;
1090 const Mat getDescriptor( int imgIdx, int localDescIdx ) const;
1091 const Mat getDescriptor( int globalDescIdx ) const;
1092 void getLocalIdx( int globalDescIdx, int& imgIdx, int& localDescIdx ) const;
1097 Mat mergedDescriptors;
1098 std::vector<int> startIdxs;
1101 // In fact the matching is implemented only by the following two methods. These methods suppose
1102 // that the class object has been trained already. Public match methods call these methods
1103 // after calling train().
1104 virtual void knnMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, int k,
1105 InputArrayOfArrays masks=noArray(), bool compactResult=false ) = 0;
1106 virtual void radiusMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, float maxDistance,
1107 InputArrayOfArrays masks=noArray(), bool compactResult=false ) = 0;
1109 static bool isPossibleMatch( InputArray mask, int queryIdx, int trainIdx );
1110 static bool isMaskedOut( InputArrayOfArrays masks, int queryIdx );
1112 static Mat clone_op( Mat m ) { return m.clone(); }
1113 void checkMasks( InputArrayOfArrays masks, int queryDescriptorsCount ) const;
1115 // Collection of descriptors from train images.
1116 std::vector<Mat> trainDescCollection;
1117 std::vector<UMat> utrainDescCollection;
1121 * Brute-force descriptor matcher.
1123 * For each descriptor in the first set, this matcher finds the closest
1124 * descriptor in the second set by trying each one.
1126 * For efficiency, BruteForceMatcher is templated on the distance metric.
1127 * For float descriptors, a common choice would be cv::L2<float>.
1129 class CV_EXPORTS_W BFMatcher : public DescriptorMatcher
1132 CV_WRAP BFMatcher( int normType=NORM_L2, bool crossCheck=false );
1133 virtual ~BFMatcher() {}
1135 virtual bool isMaskSupported() const { return true; }
1137 virtual Ptr<DescriptorMatcher> clone( bool emptyTrainData=false ) const;
1139 AlgorithmInfo* info() const;
1141 virtual void knnMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, int k,
1142 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1143 virtual void radiusMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, float maxDistance,
1144 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1152 * Flann based matcher
1154 class CV_EXPORTS_W FlannBasedMatcher : public DescriptorMatcher
1157 CV_WRAP FlannBasedMatcher( const Ptr<flann::IndexParams>& indexParams=makePtr<flann::KDTreeIndexParams>(),
1158 const Ptr<flann::SearchParams>& searchParams=makePtr<flann::SearchParams>() );
1160 virtual void add( InputArrayOfArrays descriptors );
1161 virtual void clear();
1163 // Reads matcher object from a file node
1164 virtual void read( const FileNode& );
1165 // Writes matcher object to a file storage
1166 virtual void write( FileStorage& ) const;
1168 virtual void train();
1169 virtual bool isMaskSupported() const;
1171 virtual Ptr<DescriptorMatcher> clone( bool emptyTrainData=false ) const;
1173 AlgorithmInfo* info() const;
1175 static void convertToDMatches( const DescriptorCollection& descriptors,
1176 const Mat& indices, const Mat& distances,
1177 std::vector<std::vector<DMatch> >& matches );
1179 virtual void knnMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, int k,
1180 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1181 virtual void radiusMatchImpl( InputArray queryDescriptors, std::vector<std::vector<DMatch> >& matches, float maxDistance,
1182 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1184 Ptr<flann::IndexParams> indexParams;
1185 Ptr<flann::SearchParams> searchParams;
1186 Ptr<flann::Index> flannIndex;
1188 DescriptorCollection mergedDescriptors;
1192 /****************************************************************************************\
1193 * GenericDescriptorMatcher *
1194 \****************************************************************************************/
1196 * Abstract interface for a keypoint descriptor and matcher
1198 class GenericDescriptorMatcher;
1199 typedef GenericDescriptorMatcher GenericDescriptorMatch;
1201 class CV_EXPORTS GenericDescriptorMatcher
1204 GenericDescriptorMatcher();
1205 virtual ~GenericDescriptorMatcher();
1208 * Add train collection: images and keypoints from them.
1209 * images A set of train images.
1210 * ketpoints Keypoint collection that have been detected on train images.
1212 * Keypoints for which a descriptor cannot be computed are removed. Such keypoints
1213 * must be filtered in this method befor adding keypoints to train collection "trainPointCollection".
1214 * If inheritor class need perform such prefiltering the method add() must be overloaded.
1215 * In the other class methods programmer has access to the train keypoints by a constant link.
1217 virtual void add( InputArrayOfArrays images,
1218 std::vector<std::vector<KeyPoint> >& keypoints );
1220 const std::vector<Mat>& getTrainImages() const;
1221 const std::vector<std::vector<KeyPoint> >& getTrainKeypoints() const;
1224 * Clear images and keypoints storing in train collection.
1226 virtual void clear();
1228 * Returns true if matcher supports mask to match descriptors.
1230 virtual bool isMaskSupported() = 0;
1232 * Train some inner structures (e.g. flann index or decision trees).
1233 * train() methods is run every time in matching methods. So the method implementation
1234 * should has a check whether these inner structures need be trained/retrained or not.
1236 virtual void train();
1239 * Classifies query keypoints.
1240 * queryImage The query image
1241 * queryKeypoints Keypoints from the query image
1242 * trainImage The train image
1243 * trainKeypoints Keypoints from the train image
1245 // Classify keypoints from query image under one train image.
1246 void classify( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1247 InputArray trainImage, std::vector<KeyPoint>& trainKeypoints ) const;
1248 // Classify keypoints from query image under train image collection.
1249 void classify( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints );
1252 * Group of methods to match keypoints from image pair.
1253 * Keypoints for which a descriptor cannot be computed are removed.
1254 * train() method is called here.
1256 // Find one best match for each query descriptor (if mask is empty).
1257 void match( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1258 InputArray trainImage, std::vector<KeyPoint>& trainKeypoints,
1259 std::vector<DMatch>& matches, InputArray mask=noArray() ) const;
1260 // Find k best matches for each query keypoint (in increasing order of distances).
1261 // compactResult is used when mask is not empty. If compactResult is false matches
1262 // vector will have the same size as queryDescriptors rows.
1263 // If compactResult is true matches vector will not contain matches for fully masked out query descriptors.
1264 void knnMatch( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1265 InputArray trainImage, std::vector<KeyPoint>& trainKeypoints,
1266 std::vector<std::vector<DMatch> >& matches, int k,
1267 InputArray mask=noArray(), bool compactResult=false ) const;
1268 // Find best matches for each query descriptor which have distance less than maxDistance (in increasing order of distances).
1269 void radiusMatch( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1270 InputArray trainImage, std::vector<KeyPoint>& trainKeypoints,
1271 std::vector<std::vector<DMatch> >& matches, float maxDistance,
1272 InputArray mask=noArray(), bool compactResult=false ) const;
1274 * Group of methods to match keypoints from one image to image set.
1275 * See description of similar methods for matching image pair above.
1277 void match( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1278 std::vector<DMatch>& matches, InputArrayOfArrays masks=noArray() );
1279 void knnMatch( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1280 std::vector<std::vector<DMatch> >& matches, int k,
1281 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1282 void radiusMatch(InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1283 std::vector<std::vector<DMatch> >& matches, float maxDistance,
1284 InputArrayOfArrays masks=noArray(), bool compactResult=false );
1286 // Reads matcher object from a file node
1287 virtual void read( const FileNode& fn );
1288 // Writes matcher object to a file storage
1289 virtual void write( FileStorage& fs ) const;
1291 // Return true if matching object is empty (e.g. feature detector or descriptor matcher are empty)
1292 virtual bool empty() const;
1294 // Clone the matcher. If emptyTrainData is false the method create deep copy of the object, i.e. copies
1295 // both parameters and train data. If emptyTrainData is true the method create object copy with current parameters
1296 // but with empty train data.
1297 virtual Ptr<GenericDescriptorMatcher> clone( bool emptyTrainData=false ) const = 0;
1299 static Ptr<GenericDescriptorMatcher> create( const String& genericDescritptorMatcherType,
1300 const String ¶msFilename=String() );
1303 // In fact the matching is implemented only by the following two methods. These methods suppose
1304 // that the class object has been trained already. Public match methods call these methods
1305 // after calling train().
1306 virtual void knnMatchImpl( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1307 std::vector<std::vector<DMatch> >& matches, int k,
1308 InputArrayOfArrays masks, bool compactResult ) = 0;
1309 virtual void radiusMatchImpl( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1310 std::vector<std::vector<DMatch> >& matches, float maxDistance,
1311 InputArrayOfArrays masks, bool compactResult ) = 0;
1313 * A storage for sets of keypoints together with corresponding images and class IDs
1315 class CV_EXPORTS KeyPointCollection
1318 KeyPointCollection();
1319 KeyPointCollection( const KeyPointCollection& collection );
1320 void add( const std::vector<Mat>& images, const std::vector<std::vector<KeyPoint> >& keypoints );
1323 // Returns the total number of keypoints in the collection
1324 size_t keypointCount() const;
1325 size_t imageCount() const;
1327 const std::vector<std::vector<KeyPoint> >& getKeypoints() const;
1328 const std::vector<KeyPoint>& getKeypoints( int imgIdx ) const;
1329 const KeyPoint& getKeyPoint( int imgIdx, int localPointIdx ) const;
1330 const KeyPoint& getKeyPoint( int globalPointIdx ) const;
1331 void getLocalIdx( int globalPointIdx, int& imgIdx, int& localPointIdx ) const;
1333 const std::vector<Mat>& getImages() const;
1334 const Mat& getImage( int imgIdx ) const;
1339 std::vector<Mat> images;
1340 std::vector<std::vector<KeyPoint> > keypoints;
1341 // global indices of the first points in each image, startIndices.size() = keypoints.size()
1342 std::vector<int> startIndices;
1345 static Mat clone_op( Mat m ) { return m.clone(); }
1348 KeyPointCollection trainPointCollection;
1352 /****************************************************************************************\
1353 * VectorDescriptorMatcher *
1354 \****************************************************************************************/
1357 * A class used for matching descriptors that can be described as vectors in a finite-dimensional space
1359 class VectorDescriptorMatcher;
1360 typedef VectorDescriptorMatcher VectorDescriptorMatch;
1362 class CV_EXPORTS VectorDescriptorMatcher : public GenericDescriptorMatcher
1365 VectorDescriptorMatcher( const Ptr<DescriptorExtractor>& extractor, const Ptr<DescriptorMatcher>& matcher );
1366 virtual ~VectorDescriptorMatcher();
1368 virtual void add( InputArrayOfArrays imgCollection,
1369 std::vector<std::vector<KeyPoint> >& pointCollection );
1371 virtual void clear();
1373 virtual void train();
1375 virtual bool isMaskSupported();
1377 virtual void read( const FileNode& fn );
1378 virtual void write( FileStorage& fs ) const;
1379 virtual bool empty() const;
1381 virtual Ptr<GenericDescriptorMatcher> clone( bool emptyTrainData=false ) const;
1384 virtual void knnMatchImpl( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1385 std::vector<std::vector<DMatch> >& matches, int k,
1386 InputArrayOfArrays masks, bool compactResult );
1387 virtual void radiusMatchImpl( InputArray queryImage, std::vector<KeyPoint>& queryKeypoints,
1388 std::vector<std::vector<DMatch> >& matches, float maxDistance,
1389 InputArrayOfArrays masks, bool compactResult );
1391 Ptr<DescriptorExtractor> extractor;
1392 Ptr<DescriptorMatcher> matcher;
1395 /****************************************************************************************\
1396 * Drawing functions *
1397 \****************************************************************************************/
1398 struct CV_EXPORTS DrawMatchesFlags
1400 enum{ DEFAULT = 0, // Output image matrix will be created (Mat::create),
1401 // i.e. existing memory of output image may be reused.
1402 // Two source image, matches and single keypoints will be drawn.
1403 // For each keypoint only the center point will be drawn (without
1404 // the circle around keypoint with keypoint size and orientation).
1405 DRAW_OVER_OUTIMG = 1, // Output image matrix will not be created (Mat::create).
1406 // Matches will be drawn on existing content of output image.
1407 NOT_DRAW_SINGLE_POINTS = 2, // Single keypoints will not be drawn.
1408 DRAW_RICH_KEYPOINTS = 4 // For each keypoint the circle around keypoint with keypoint size and
1409 // orientation will be drawn.
1414 CV_EXPORTS_W void drawKeypoints( InputArray image, const std::vector<KeyPoint>& keypoints, InputOutputArray outImage,
1415 const Scalar& color=Scalar::all(-1), int flags=DrawMatchesFlags::DEFAULT );
1417 // Draws matches of keypints from two images on output image.
1418 CV_EXPORTS_W void drawMatches( InputArray img1, const std::vector<KeyPoint>& keypoints1,
1419 InputArray img2, const std::vector<KeyPoint>& keypoints2,
1420 const std::vector<DMatch>& matches1to2, InputOutputArray outImg,
1421 const Scalar& matchColor=Scalar::all(-1), const Scalar& singlePointColor=Scalar::all(-1),
1422 const std::vector<char>& matchesMask=std::vector<char>(), int flags=DrawMatchesFlags::DEFAULT );
1424 CV_EXPORTS_AS(drawMatchesKnn) void drawMatches( InputArray img1, const std::vector<KeyPoint>& keypoints1,
1425 InputArray img2, const std::vector<KeyPoint>& keypoints2,
1426 const std::vector<std::vector<DMatch> >& matches1to2, InputOutputArray outImg,
1427 const Scalar& matchColor=Scalar::all(-1), const Scalar& singlePointColor=Scalar::all(-1),
1428 const std::vector<std::vector<char> >& matchesMask=std::vector<std::vector<char> >(), int flags=DrawMatchesFlags::DEFAULT );
1430 /****************************************************************************************\
1431 * Functions to evaluate the feature detectors and [generic] descriptor extractors *
1432 \****************************************************************************************/
1434 CV_EXPORTS void evaluateFeatureDetector( const Mat& img1, const Mat& img2, const Mat& H1to2,
1435 std::vector<KeyPoint>* keypoints1, std::vector<KeyPoint>* keypoints2,
1436 float& repeatability, int& correspCount,
1437 const Ptr<FeatureDetector>& fdetector=Ptr<FeatureDetector>() );
1439 CV_EXPORTS void computeRecallPrecisionCurve( const std::vector<std::vector<DMatch> >& matches1to2,
1440 const std::vector<std::vector<uchar> >& correctMatches1to2Mask,
1441 std::vector<Point2f>& recallPrecisionCurve );
1443 CV_EXPORTS float getRecall( const std::vector<Point2f>& recallPrecisionCurve, float l_precision );
1444 CV_EXPORTS int getNearestPoint( const std::vector<Point2f>& recallPrecisionCurve, float l_precision );
1446 CV_EXPORTS void evaluateGenericDescriptorMatcher( const Mat& img1, const Mat& img2, const Mat& H1to2,
1447 std::vector<KeyPoint>& keypoints1, std::vector<KeyPoint>& keypoints2,
1448 std::vector<std::vector<DMatch> >* matches1to2, std::vector<std::vector<uchar> >* correctMatches1to2Mask,
1449 std::vector<Point2f>& recallPrecisionCurve,
1450 const Ptr<GenericDescriptorMatcher>& dmatch=Ptr<GenericDescriptorMatcher>() );
1453 /****************************************************************************************\
1454 * Bag of visual words *
1455 \****************************************************************************************/
1457 * Abstract base class for training of a 'bag of visual words' vocabulary from a set of descriptors
1459 class CV_EXPORTS BOWTrainer
1463 virtual ~BOWTrainer();
1465 void add( const Mat& descriptors );
1466 const std::vector<Mat>& getDescriptors() const;
1467 int descriptorsCount() const;
1469 virtual void clear();
1472 * Train visual words vocabulary, that is cluster training descriptors and
1473 * compute cluster centers.
1474 * Returns cluster centers.
1476 * descriptors Training descriptors computed on images keypoints.
1478 virtual Mat cluster() const = 0;
1479 virtual Mat cluster( const Mat& descriptors ) const = 0;
1482 std::vector<Mat> descriptors;
1487 * This is BOWTrainer using cv::kmeans to get vocabulary.
1489 class CV_EXPORTS BOWKMeansTrainer : public BOWTrainer
1492 BOWKMeansTrainer( int clusterCount, const TermCriteria& termcrit=TermCriteria(),
1493 int attempts=3, int flags=KMEANS_PP_CENTERS );
1494 virtual ~BOWKMeansTrainer();
1496 // Returns trained vocabulary (i.e. cluster centers).
1497 virtual Mat cluster() const;
1498 virtual Mat cluster( const Mat& descriptors ) const;
1503 TermCriteria termcrit;
1509 * Class to compute image descriptor using bag of visual words.
1511 class CV_EXPORTS BOWImgDescriptorExtractor
1514 BOWImgDescriptorExtractor( const Ptr<DescriptorExtractor>& dextractor,
1515 const Ptr<DescriptorMatcher>& dmatcher );
1516 BOWImgDescriptorExtractor( const Ptr<DescriptorMatcher>& dmatcher );
1517 virtual ~BOWImgDescriptorExtractor();
1519 void setVocabulary( const Mat& vocabulary );
1520 const Mat& getVocabulary() const;
1521 void compute( InputArray image, std::vector<KeyPoint>& keypoints, OutputArray imgDescriptor,
1522 std::vector<std::vector<int> >* pointIdxsOfClusters=0, Mat* descriptors=0 );
1523 void compute( InputArray keypointDescriptors, OutputArray imgDescriptor,
1524 std::vector<std::vector<int> >* pointIdxsOfClusters=0 );
1525 // compute() is not constant because DescriptorMatcher::match is not constant
1527 int descriptorSize() const;
1528 int descriptorType() const;
1532 Ptr<DescriptorExtractor> dextractor;
1533 Ptr<DescriptorMatcher> dmatcher;
1536 } /* namespace cv */