• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*M///////////////////////////////////////////////////////////////////////////////////////
2 //
3 //  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4 //
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.
8 //
9 //
10 //                        Intel License Agreement
11 //
12 // Copyright (C) 2000, Intel Corporation, all rights reserved.
13 // Third party copyrights are property of their respective owners.
14 //
15 // Redistribution and use in source and binary forms, with or without modification,
16 // are permitted provided that the following conditions are met:
17 //
18 //   * Redistribution's of source code must retain the above copyright notice,
19 //     this list of conditions and the following disclaimer.
20 //
21 //   * Redistribution's in binary form must reproduce the above copyright notice,
22 //     this list of conditions and the following disclaimer in the documentation
23 //     and/or other materials provided with the distribution.
24 //
25 //   * The name of Intel Corporation may not be used to endorse or promote products
26 //     derived from this software without specific prior written permission.
27 //
28 // This software is provided by the copyright holders and contributors "as is" and
29 // any express or implied warranties, including, but not limited to, the implied
30 // warranties of merchantability and fitness for a particular purpose are disclaimed.
31 // In no event shall the Intel Corporation or contributors be liable for any direct,
32 // indirect, incidental, special, exemplary, or consequential damages
33 // (including, but not limited to, procurement of substitute goods or services;
34 // loss of use, data, or profits; or business interruption) however caused
35 // and on any theory of liability, whether in contract, strict liability,
36 // or tort (including negligence or otherwise) arising in any way out of
37 // the use of this software, even if advised of the possibility of such damage.
38 //
39 //M*/
40 
41 #include "_ml.h"
42 
43 
CvStatModel()44 CvStatModel::CvStatModel()
45 {
46     default_model_name = "my_stat_model";
47 }
48 
49 
~CvStatModel()50 CvStatModel::~CvStatModel()
51 {
52     clear();
53 }
54 
55 
clear()56 void CvStatModel::clear()
57 {
58 }
59 
60 
save(const char * filename,const char * name)61 void CvStatModel::save( const char* filename, const char* name )
62 {
63     CvFileStorage* fs = 0;
64 
65     CV_FUNCNAME( "CvStatModel::save" );
66 
67     __BEGIN__;
68 
69     CV_CALL( fs = cvOpenFileStorage( filename, 0, CV_STORAGE_WRITE ));
70     if( !fs )
71         CV_ERROR( CV_StsError, "Could not open the file storage. Check the path and permissions" );
72 
73     write( fs, name ? name : default_model_name );
74 
75     __END__;
76 
77     cvReleaseFileStorage( &fs );
78 }
79 
80 
load(const char * filename,const char * name)81 void CvStatModel::load( const char* filename, const char* name )
82 {
83     CvFileStorage* fs = 0;
84 
85     CV_FUNCNAME( "CvStatModel::load" );
86 
87     __BEGIN__;
88 
89     CvFileNode* model_node = 0;
90 
91     CV_CALL( fs = cvOpenFileStorage( filename, 0, CV_STORAGE_READ ));
92     if( !fs )
93         EXIT;
94 
95     if( name )
96         model_node = cvGetFileNodeByName( fs, 0, name );
97     else
98     {
99         CvFileNode* root = cvGetRootFileNode( fs );
100         if( root->data.seq->total > 0 )
101             model_node = (CvFileNode*)cvGetSeqElem( root->data.seq, 0 );
102     }
103 
104     read( fs, model_node );
105 
106     __END__;
107 
108     cvReleaseFileStorage( &fs );
109 }
110 
111 
write(CvFileStorage *,const char *)112 void CvStatModel::write( CvFileStorage*, const char* )
113 {
114     OPENCV_ERROR( CV_StsNotImplemented, "CvStatModel::write", "" );
115 }
116 
117 
read(CvFileStorage *,CvFileNode *)118 void CvStatModel::read( CvFileStorage*, CvFileNode* )
119 {
120     OPENCV_ERROR( CV_StsNotImplemented, "CvStatModel::read", "" );
121 }
122 
123 
124 /* Calculates upper triangular matrix S, where A is a symmetrical matrix A=S'*S */
cvChol(CvMat * A,CvMat * S)125 CV_IMPL void cvChol( CvMat* A, CvMat* S )
126 {
127     int dim = A->rows;
128 
129     int i, j, k;
130     float sum;
131 
132     for( i = 0; i < dim; i++ )
133     {
134         for( j = 0; j < i; j++ )
135             CV_MAT_ELEM(*S, float, i, j) = 0;
136 
137         sum = 0;
138         for( k = 0; k < i; k++ )
139             sum += CV_MAT_ELEM(*S, float, k, i) * CV_MAT_ELEM(*S, float, k, i);
140 
141         CV_MAT_ELEM(*S, float, i, i) = (float)sqrt(CV_MAT_ELEM(*A, float, i, i) - sum);
142 
143         for( j = i + 1; j < dim; j++ )
144         {
145             sum = 0;
146             for( k = 0; k < i; k++ )
147                 sum += CV_MAT_ELEM(*S, float, k, i) * CV_MAT_ELEM(*S, float, k, j);
148 
149             CV_MAT_ELEM(*S, float, i, j) =
150                 (CV_MAT_ELEM(*A, float, i, j) - sum) / CV_MAT_ELEM(*S, float, i, i);
151 
152         }
153     }
154 }
155 
156 /* Generates <sample> from multivariate normal distribution, where <mean> - is an
157    average row vector, <cov> - symmetric covariation matrix */
cvRandMVNormal(CvMat * mean,CvMat * cov,CvMat * sample,CvRNG * rng)158 CV_IMPL void cvRandMVNormal( CvMat* mean, CvMat* cov, CvMat* sample, CvRNG* rng )
159 {
160     int dim = sample->cols;
161     int amount = sample->rows;
162 
163     CvRNG state = rng ? *rng : cvRNG(time(0));
164     cvRandArr(&state, sample, CV_RAND_NORMAL, cvScalarAll(0), cvScalarAll(1) );
165 
166     CvMat* utmat = cvCreateMat(dim, dim, sample->type);
167     CvMat* vect = cvCreateMatHeader(1, dim, sample->type);
168 
169     cvChol(cov, utmat);
170 
171     int i;
172     for( i = 0; i < amount; i++ )
173     {
174         cvGetRow(sample, vect, i);
175         cvMatMulAdd(vect, utmat, mean, vect);
176     }
177 
178     cvReleaseMat(&vect);
179     cvReleaseMat(&utmat);
180 }
181 
182 
183 /* Generates <sample> of <amount> points from a discrete variate xi,
184    where Pr{xi = k} == probs[k], 0 < k < len - 1. */
cvRandSeries(float probs[],int len,int sample[],int amount)185 CV_IMPL void cvRandSeries( float probs[], int len, int sample[], int amount )
186 {
187     CvMat* univals = cvCreateMat(1, amount, CV_32FC1);
188     float* knots = (float*)cvAlloc( len * sizeof(float) );
189 
190     int i, j;
191 
192     CvRNG state = cvRNG(-1);
193     cvRandArr(&state, univals, CV_RAND_UNI, cvScalarAll(0), cvScalarAll(1) );
194 
195     knots[0] = probs[0];
196     for( i = 1; i < len; i++ )
197         knots[i] = knots[i - 1] + probs[i];
198 
199     for( i = 0; i < amount; i++ )
200         for( j = 0; j < len; j++ )
201         {
202             if ( CV_MAT_ELEM(*univals, float, 0, i) <= knots[j] )
203             {
204                 sample[i] = j;
205                 break;
206             }
207         }
208 
209     cvFree(&knots);
210 }
211 
212 /* Generates <sample> from gaussian mixture distribution */
cvRandGaussMixture(CvMat * means[],CvMat * covs[],float weights[],int clsnum,CvMat * sample,CvMat * sampClasses)213 CV_IMPL void cvRandGaussMixture( CvMat* means[],
214                                  CvMat* covs[],
215                                  float weights[],
216                                  int clsnum,
217                                  CvMat* sample,
218                                  CvMat* sampClasses )
219 {
220     int dim = sample->cols;
221     int amount = sample->rows;
222 
223     int i, clss;
224 
225     int* sample_clsnum = (int*)cvAlloc( amount * sizeof(int) );
226     CvMat** utmats = (CvMat**)cvAlloc( clsnum * sizeof(CvMat*) );
227     CvMat* vect = cvCreateMatHeader(1, dim, CV_32FC1);
228 
229     CvMat* classes;
230     if( sampClasses )
231         classes = sampClasses;
232     else
233         classes = cvCreateMat(1, amount, CV_32FC1);
234 
235     CvRNG state = cvRNG(-1);
236     cvRandArr(&state, sample, CV_RAND_NORMAL, cvScalarAll(0), cvScalarAll(1));
237 
238     cvRandSeries(weights, clsnum, sample_clsnum, amount);
239 
240     for( i = 0; i < clsnum; i++ )
241     {
242         utmats[i] = cvCreateMat(dim, dim, CV_32FC1);
243         cvChol(covs[i], utmats[i]);
244     }
245 
246     for( i = 0; i < amount; i++ )
247     {
248         CV_MAT_ELEM(*classes, float, 0, i) = (float)sample_clsnum[i];
249         cvGetRow(sample, vect, i);
250         clss = sample_clsnum[i];
251         cvMatMulAdd(vect, utmats[clss], means[clss], vect);
252     }
253 
254     if( !sampClasses )
255         cvReleaseMat(&classes);
256     for( i = 0; i < clsnum; i++ )
257         cvReleaseMat(&utmats[i]);
258     cvFree(&utmats);
259     cvFree(&sample_clsnum);
260     cvReleaseMat(&vect);
261 }
262 
263 
icvGenerateRandomClusterCenters(int seed,const CvMat * data,int num_of_clusters,CvMat * _centers)264 CvMat* icvGenerateRandomClusterCenters ( int seed, const CvMat* data,
265                                          int num_of_clusters, CvMat* _centers )
266 {
267     CvMat* centers = _centers;
268 
269     CV_FUNCNAME("icvGenerateRandomClusterCenters");
270     __BEGIN__;
271 
272     CvRNG rng;
273     CvMat data_comp, centers_comp;
274     CvPoint minLoc, maxLoc; // Not used, just for function "cvMinMaxLoc"
275     double minVal, maxVal;
276     int i;
277     int dim = data ? data->cols : 0;
278 
279     if( ICV_IS_MAT_OF_TYPE(data, CV_32FC1) )
280     {
281         if( _centers && !ICV_IS_MAT_OF_TYPE (_centers, CV_32FC1) )
282         {
283             CV_ERROR(CV_StsBadArg,"");
284         }
285         else if( !_centers )
286             CV_CALL(centers = cvCreateMat (num_of_clusters, dim, CV_32FC1));
287     }
288     else if( ICV_IS_MAT_OF_TYPE(data, CV_64FC1) )
289     {
290         if( _centers && !ICV_IS_MAT_OF_TYPE (_centers, CV_64FC1) )
291         {
292             CV_ERROR(CV_StsBadArg,"");
293         }
294         else if( !_centers )
295             CV_CALL(centers = cvCreateMat (num_of_clusters, dim, CV_64FC1));
296     }
297     else
298         CV_ERROR (CV_StsBadArg,"");
299 
300     if( num_of_clusters < 1 )
301         CV_ERROR (CV_StsBadArg,"");
302 
303     rng = cvRNG(seed);
304     for (i = 0; i < dim; i++)
305     {
306         CV_CALL(cvGetCol (data, &data_comp, i));
307         CV_CALL(cvMinMaxLoc (&data_comp, &minVal, &maxVal, &minLoc, &maxLoc));
308         CV_CALL(cvGetCol (centers, &centers_comp, i));
309         CV_CALL(cvRandArr (&rng, &centers_comp, CV_RAND_UNI, cvScalarAll(minVal), cvScalarAll(maxVal)));
310     }
311 
312     __END__;
313 
314     if( (cvGetErrStatus () < 0) || (centers != _centers) )
315         cvReleaseMat (&centers);
316 
317     return _centers ? _centers : centers;
318 } // end of icvGenerateRandomClusterCenters
319 
320 // By S. Dilman - begin -
321 
322 #define ICV_RAND_MAX    4294967296 // == 2^32
323 
cvRandRoundUni(CvMat * center,float radius_small,float radius_large,CvMat * desired_matrix,CvRNG * rng_state_ptr)324 CV_IMPL void cvRandRoundUni (CvMat* center,
325                              float radius_small,
326                              float radius_large,
327                              CvMat* desired_matrix,
328                              CvRNG* rng_state_ptr)
329 {
330     float rad, norm, coefficient;
331     int dim, size, i, j;
332     CvMat *cov, sample;
333     CvRNG rng_local;
334 
335     CV_FUNCNAME("cvRandRoundUni");
336     __BEGIN__
337 
338     rng_local = *rng_state_ptr;
339 
340     CV_ASSERT ((radius_small >= 0) &&
341                (radius_large > 0) &&
342                (radius_small <= radius_large));
343     CV_ASSERT (center && desired_matrix && rng_state_ptr);
344     CV_ASSERT (center->rows == 1);
345     CV_ASSERT (center->cols == desired_matrix->cols);
346 
347     dim = desired_matrix->cols;
348     size = desired_matrix->rows;
349     cov = cvCreateMat (dim, dim, CV_32FC1);
350     cvSetIdentity (cov);
351     cvRandMVNormal (center, cov, desired_matrix, &rng_local);
352 
353     for (i = 0; i < size; i++)
354     {
355         rad = (float)(cvRandReal(&rng_local)*(radius_large - radius_small) + radius_small);
356         cvGetRow (desired_matrix, &sample, i);
357         norm = (float) cvNorm (&sample, 0, CV_L2);
358         coefficient = rad / norm;
359         for (j = 0; j < dim; j++)
360              CV_MAT_ELEM (sample, float, 0, j) *= coefficient;
361     }
362 
363     __END__
364 
365 }
366 
367 // By S. Dilman - end -
368 
369 /* Aij <- Aji for i > j if lower_to_upper != 0
370               for i < j if lower_to_upper = 0 */
cvCompleteSymm(CvMat * matrix,int lower_to_upper)371 void cvCompleteSymm( CvMat* matrix, int lower_to_upper )
372 {
373     CV_FUNCNAME("cvCompleteSymm");
374 
375     __BEGIN__;
376 
377     int rows, cols;
378     int i, j;
379     int step;
380 
381     if( !CV_IS_MAT(matrix))
382         CV_ERROR(CV_StsBadArg, "Invalid matrix argument");
383 
384     rows = matrix->rows;
385     cols = matrix->cols;
386     step = matrix->step / CV_ELEM_SIZE(matrix->type);
387 
388     switch(CV_MAT_TYPE(matrix->type))
389     {
390     case CV_32FC1:
391         {
392         float* dst = matrix->data.fl;
393         if( !lower_to_upper )
394             for( i = 1; i < rows; i++ )
395             {
396                 const float* src = (const float*)(matrix->data.fl + i);
397                 dst += step;
398                 for( j = 0; j < i; j++, src += step )
399                     dst[j] = src[0];
400             }
401         else
402             for( i = 0; i < rows-1; i++, dst += step )
403             {
404                 const float* src = (const float*)(matrix->data.fl + (i+1)*step + i);
405                 for( j = i+1; j < cols; j++, src += step )
406                     dst[j] = src[0];
407             }
408         }
409         break;
410     case CV_64FC1:
411         {
412         double* dst = matrix->data.db;
413         if( !lower_to_upper )
414             for( i = 1; i < rows; i++ )
415             {
416                 const double* src = (const double*)(matrix->data.db + i);
417                 dst += step;
418                 for( j = 0; j < i; j++, src += step )
419                     dst[j] = src[0];
420             }
421         else
422             for( i = 0; i < rows-1; i++, dst += step )
423             {
424                 const double* src = (const double*)(matrix->data.db + (i+1)*step + i);
425                 for( j = i+1; j < cols; j++, src += step )
426                     dst[j] = src[0];
427             }
428         }
429         break;
430     }
431 
432     __END__;
433 }
434 
435 
436 static int CV_CDECL
icvCmpIntegers(const void * a,const void * b)437 icvCmpIntegers( const void* a, const void* b )
438 {
439     return *(const int*)a - *(const int*)b;
440 }
441 
442 
443 static int CV_CDECL
icvCmpIntegersPtr(const void * _a,const void * _b)444 icvCmpIntegersPtr( const void* _a, const void* _b )
445 {
446     int a = **(const int**)_a;
447     int b = **(const int**)_b;
448     return (a < b ? -1 : 0)|(a > b);
449 }
450 
451 
icvCmpSparseVecElems(const void * a,const void * b)452 static int icvCmpSparseVecElems( const void* a, const void* b )
453 {
454     return ((CvSparseVecElem32f*)a)->idx - ((CvSparseVecElem32f*)b)->idx;
455 }
456 
457 
458 CvMat*
cvPreprocessIndexArray(const CvMat * idx_arr,int data_arr_size,bool check_for_duplicates)459 cvPreprocessIndexArray( const CvMat* idx_arr, int data_arr_size, bool check_for_duplicates )
460 {
461     CvMat* idx = 0;
462 
463     CV_FUNCNAME( "cvPreprocessIndexArray" );
464 
465     __BEGIN__;
466 
467     int i, idx_total, idx_selected = 0, step, type, prev = INT_MIN, is_sorted = 1;
468     uchar* srcb = 0;
469     int* srci = 0;
470     int* dsti;
471 
472     if( !CV_IS_MAT(idx_arr) )
473         CV_ERROR( CV_StsBadArg, "Invalid index array" );
474 
475     if( idx_arr->rows != 1 && idx_arr->cols != 1 )
476         CV_ERROR( CV_StsBadSize, "the index array must be 1-dimensional" );
477 
478     idx_total = idx_arr->rows + idx_arr->cols - 1;
479     srcb = idx_arr->data.ptr;
480     srci = idx_arr->data.i;
481 
482     type = CV_MAT_TYPE(idx_arr->type);
483     step = CV_IS_MAT_CONT(idx_arr->type) ? 1 : idx_arr->step/CV_ELEM_SIZE(type);
484 
485     switch( type )
486     {
487     case CV_8UC1:
488     case CV_8SC1:
489         // idx_arr is array of 1's and 0's -
490         // i.e. it is a mask of the selected components
491         if( idx_total != data_arr_size )
492             CV_ERROR( CV_StsUnmatchedSizes,
493             "Component mask should contain as many elements as the total number of input variables" );
494 
495         for( i = 0; i < idx_total; i++ )
496             idx_selected += srcb[i*step] != 0;
497 
498         if( idx_selected == 0 )
499             CV_ERROR( CV_StsOutOfRange, "No components/input_variables is selected!" );
500 
501         if( idx_selected == idx_total )
502             EXIT;
503         break;
504     case CV_32SC1:
505         // idx_arr is array of integer indices of selected components
506         if( idx_total > data_arr_size )
507             CV_ERROR( CV_StsOutOfRange,
508             "index array may not contain more elements than the total number of input variables" );
509         idx_selected = idx_total;
510         // check if sorted already
511         for( i = 0; i < idx_total; i++ )
512         {
513             int val = srci[i*step];
514             if( val >= prev )
515             {
516                 is_sorted = 0;
517                 break;
518             }
519             prev = val;
520         }
521         break;
522     default:
523         CV_ERROR( CV_StsUnsupportedFormat, "Unsupported index array data type "
524                                            "(it should be 8uC1, 8sC1 or 32sC1)" );
525     }
526 
527     CV_CALL( idx = cvCreateMat( 1, idx_selected, CV_32SC1 ));
528     dsti = idx->data.i;
529 
530     if( type < CV_32SC1 )
531     {
532         for( i = 0; i < idx_total; i++ )
533             if( srcb[i*step] )
534                 *dsti++ = i;
535     }
536     else
537     {
538         for( i = 0; i < idx_total; i++ )
539             dsti[i] = srci[i*step];
540 
541         if( !is_sorted )
542             qsort( dsti, idx_total, sizeof(dsti[0]), icvCmpIntegers );
543 
544         if( dsti[0] < 0 || dsti[idx_total-1] >= data_arr_size )
545             CV_ERROR( CV_StsOutOfRange, "the index array elements are out of range" );
546 
547         if( check_for_duplicates )
548         {
549             for( i = 1; i < idx_total; i++ )
550                 if( dsti[i] <= dsti[i-1] )
551                     CV_ERROR( CV_StsBadArg, "There are duplicated index array elements" );
552         }
553     }
554 
555     __END__;
556 
557     if( cvGetErrStatus() < 0 )
558         cvReleaseMat( &idx );
559 
560     return idx;
561 }
562 
563 
564 CvMat*
cvPreprocessVarType(const CvMat * var_type,const CvMat * var_idx,int var_all,int * response_type)565 cvPreprocessVarType( const CvMat* var_type, const CvMat* var_idx,
566                      int var_all, int* response_type )
567 {
568     CvMat* out_var_type = 0;
569     CV_FUNCNAME( "cvPreprocessVarType" );
570 
571     if( response_type )
572         *response_type = -1;
573 
574     __BEGIN__;
575 
576     int i, tm_size, tm_step;
577     int* map = 0;
578     const uchar* src;
579     uchar* dst;
580     int var_count = var_all;
581 
582     if( !CV_IS_MAT(var_type) )
583         CV_ERROR( var_type ? CV_StsBadArg : CV_StsNullPtr, "Invalid or absent var_type array" );
584 
585     if( var_type->rows != 1 && var_type->cols != 1 )
586         CV_ERROR( CV_StsBadSize, "var_type array must be 1-dimensional" );
587 
588     if( !CV_IS_MASK_ARR(var_type))
589         CV_ERROR( CV_StsUnsupportedFormat, "type mask must be 8uC1 or 8sC1 array" );
590 
591     tm_size = var_type->rows + var_type->cols - 1;
592     tm_step = var_type->step ? var_type->step/CV_ELEM_SIZE(var_type->type) : 1;
593 
594     if( /*tm_size != var_count &&*/ tm_size != var_count + 1 )
595         CV_ERROR( CV_StsBadArg,
596         "type mask must be of <input var count> + 1 size" );
597 
598     if( response_type && tm_size > var_count )
599         *response_type = var_type->data.ptr[var_count*tm_step] != 0;
600 
601     if( var_idx )
602     {
603         if( !CV_IS_MAT(var_idx) || CV_MAT_TYPE(var_idx->type) != CV_32SC1 ||
604             var_idx->rows != 1 && var_idx->cols != 1 || !CV_IS_MAT_CONT(var_idx->type) )
605             CV_ERROR( CV_StsBadArg, "var index array should be continuous 1-dimensional integer vector" );
606         if( var_idx->rows + var_idx->cols - 1 > var_count )
607             CV_ERROR( CV_StsBadSize, "var index array is too large" );
608         map = var_idx->data.i;
609         var_count = var_idx->rows + var_idx->cols - 1;
610     }
611 
612     CV_CALL( out_var_type = cvCreateMat( 1, var_count, CV_8UC1 ));
613     src = var_type->data.ptr;
614     dst = out_var_type->data.ptr;
615 
616     for( i = 0; i < var_count; i++ )
617     {
618         int idx = map ? map[i] : i;
619         assert( (unsigned)idx < (unsigned)tm_size );
620         dst[i] = (uchar)(src[idx*tm_step] != 0);
621     }
622 
623     __END__;
624 
625     return out_var_type;
626 }
627 
628 
629 CvMat*
cvPreprocessOrderedResponses(const CvMat * responses,const CvMat * sample_idx,int sample_all)630 cvPreprocessOrderedResponses( const CvMat* responses, const CvMat* sample_idx, int sample_all )
631 {
632     CvMat* out_responses = 0;
633 
634     CV_FUNCNAME( "cvPreprocessOrderedResponses" );
635 
636     __BEGIN__;
637 
638     int i, r_type, r_step;
639     const int* map = 0;
640     float* dst;
641     int sample_count = sample_all;
642 
643     if( !CV_IS_MAT(responses) )
644         CV_ERROR( CV_StsBadArg, "Invalid response array" );
645 
646     if( responses->rows != 1 && responses->cols != 1 )
647         CV_ERROR( CV_StsBadSize, "Response array must be 1-dimensional" );
648 
649     if( responses->rows + responses->cols - 1 != sample_count )
650         CV_ERROR( CV_StsUnmatchedSizes,
651         "Response array must contain as many elements as the total number of samples" );
652 
653     r_type = CV_MAT_TYPE(responses->type);
654     if( r_type != CV_32FC1 && r_type != CV_32SC1 )
655         CV_ERROR( CV_StsUnsupportedFormat, "Unsupported response type" );
656 
657     r_step = responses->step ? responses->step / CV_ELEM_SIZE(responses->type) : 1;
658 
659     if( r_type == CV_32FC1 && CV_IS_MAT_CONT(responses->type) && !sample_idx )
660     {
661         out_responses = (CvMat*)responses;
662         EXIT;
663     }
664 
665     if( sample_idx )
666     {
667         if( !CV_IS_MAT(sample_idx) || CV_MAT_TYPE(sample_idx->type) != CV_32SC1 ||
668             sample_idx->rows != 1 && sample_idx->cols != 1 || !CV_IS_MAT_CONT(sample_idx->type) )
669             CV_ERROR( CV_StsBadArg, "sample index array should be continuous 1-dimensional integer vector" );
670         if( sample_idx->rows + sample_idx->cols - 1 > sample_count )
671             CV_ERROR( CV_StsBadSize, "sample index array is too large" );
672         map = sample_idx->data.i;
673         sample_count = sample_idx->rows + sample_idx->cols - 1;
674     }
675 
676     CV_CALL( out_responses = cvCreateMat( 1, sample_count, CV_32FC1 ));
677 
678     dst = out_responses->data.fl;
679     if( r_type == CV_32FC1 )
680     {
681         const float* src = responses->data.fl;
682         for( i = 0; i < sample_count; i++ )
683         {
684             int idx = map ? map[i] : i;
685             assert( (unsigned)idx < (unsigned)sample_all );
686             dst[i] = src[idx*r_step];
687         }
688     }
689     else
690     {
691         const int* src = responses->data.i;
692         for( i = 0; i < sample_count; i++ )
693         {
694             int idx = map ? map[i] : i;
695             assert( (unsigned)idx < (unsigned)sample_all );
696             dst[i] = (float)src[idx*r_step];
697         }
698     }
699 
700     __END__;
701 
702     return out_responses;
703 }
704 
705 CvMat*
cvPreprocessCategoricalResponses(const CvMat * responses,const CvMat * sample_idx,int sample_all,CvMat ** out_response_map,CvMat ** class_counts)706 cvPreprocessCategoricalResponses( const CvMat* responses,
707     const CvMat* sample_idx, int sample_all,
708     CvMat** out_response_map, CvMat** class_counts )
709 {
710     CvMat* out_responses = 0;
711     int** response_ptr = 0;
712 
713     CV_FUNCNAME( "cvPreprocessCategoricalResponses" );
714 
715     if( out_response_map )
716         *out_response_map = 0;
717 
718     if( class_counts )
719         *class_counts = 0;
720 
721     __BEGIN__;
722 
723     int i, r_type, r_step;
724     int cls_count = 1, prev_cls, prev_i;
725     const int* map = 0;
726     const int* srci;
727     const float* srcfl;
728     int* dst;
729     int* cls_map;
730     int* cls_counts = 0;
731     int sample_count = sample_all;
732 
733     if( !CV_IS_MAT(responses) )
734         CV_ERROR( CV_StsBadArg, "Invalid response array" );
735 
736     if( responses->rows != 1 && responses->cols != 1 )
737         CV_ERROR( CV_StsBadSize, "Response array must be 1-dimensional" );
738 
739     if( responses->rows + responses->cols - 1 != sample_count )
740         CV_ERROR( CV_StsUnmatchedSizes,
741         "Response array must contain as many elements as the total number of samples" );
742 
743     r_type = CV_MAT_TYPE(responses->type);
744     if( r_type != CV_32FC1 && r_type != CV_32SC1 )
745         CV_ERROR( CV_StsUnsupportedFormat, "Unsupported response type" );
746 
747     r_step = responses->step ? responses->step / CV_ELEM_SIZE(responses->type) : 1;
748 
749     if( sample_idx )
750     {
751         if( !CV_IS_MAT(sample_idx) || CV_MAT_TYPE(sample_idx->type) != CV_32SC1 ||
752             sample_idx->rows != 1 && sample_idx->cols != 1 || !CV_IS_MAT_CONT(sample_idx->type) )
753             CV_ERROR( CV_StsBadArg, "sample index array should be continuous 1-dimensional integer vector" );
754         if( sample_idx->rows + sample_idx->cols - 1 > sample_count )
755             CV_ERROR( CV_StsBadSize, "sample index array is too large" );
756         map = sample_idx->data.i;
757         sample_count = sample_idx->rows + sample_idx->cols - 1;
758     }
759 
760     CV_CALL( out_responses = cvCreateMat( 1, sample_count, CV_32SC1 ));
761 
762     if( !out_response_map )
763         CV_ERROR( CV_StsNullPtr, "out_response_map pointer is NULL" );
764 
765     CV_CALL( response_ptr = (int**)cvAlloc( sample_count*sizeof(response_ptr[0])));
766 
767     srci = responses->data.i;
768     srcfl = responses->data.fl;
769     dst = out_responses->data.i;
770 
771     for( i = 0; i < sample_count; i++ )
772     {
773         int idx = map ? map[i] : i;
774         assert( (unsigned)idx < (unsigned)sample_all );
775         if( r_type == CV_32SC1 )
776             dst[i] = srci[idx*r_step];
777         else
778         {
779             float rf = srcfl[idx*r_step];
780             int ri = cvRound(rf);
781             if( ri != rf )
782             {
783                 char buf[100];
784                 sprintf( buf, "response #%d is not integral", idx );
785                 CV_ERROR( CV_StsBadArg, buf );
786             }
787             dst[i] = ri;
788         }
789         response_ptr[i] = dst + i;
790     }
791 
792     qsort( response_ptr, sample_count, sizeof(int*), icvCmpIntegersPtr );
793 
794     // count the classes
795     for( i = 1; i < sample_count; i++ )
796         cls_count += *response_ptr[i] != *response_ptr[i-1];
797 
798     if( cls_count < 2 )
799         CV_ERROR( CV_StsBadArg, "There is only a single class" );
800 
801     CV_CALL( *out_response_map = cvCreateMat( 1, cls_count, CV_32SC1 ));
802 
803     if( class_counts )
804     {
805         CV_CALL( *class_counts = cvCreateMat( 1, cls_count, CV_32SC1 ));
806         cls_counts = (*class_counts)->data.i;
807     }
808 
809     // compact the class indices and build the map
810     prev_cls = ~*response_ptr[0];
811     cls_count = -1;
812     cls_map = (*out_response_map)->data.i;
813 
814     for( i = 0, prev_i = -1; i < sample_count; i++ )
815     {
816         int cur_cls = *response_ptr[i];
817         if( cur_cls != prev_cls )
818         {
819             if( cls_counts && cls_count >= 0 )
820                 cls_counts[cls_count] = i - prev_i;
821             cls_map[++cls_count] = prev_cls = cur_cls;
822             prev_i = i;
823         }
824         *response_ptr[i] = cls_count;
825     }
826 
827     if( cls_counts )
828         cls_counts[cls_count] = i - prev_i;
829 
830     __END__;
831 
832     cvFree( &response_ptr );
833 
834     return out_responses;
835 }
836 
837 
838 const float**
cvGetTrainSamples(const CvMat * train_data,int tflag,const CvMat * var_idx,const CvMat * sample_idx,int * _var_count,int * _sample_count,bool always_copy_data)839 cvGetTrainSamples( const CvMat* train_data, int tflag,
840                    const CvMat* var_idx, const CvMat* sample_idx,
841                    int* _var_count, int* _sample_count,
842                    bool always_copy_data )
843 {
844     float** samples = 0;
845 
846     CV_FUNCNAME( "cvGetTrainSamples" );
847 
848     __BEGIN__;
849 
850     int i, j, var_count, sample_count, s_step, v_step;
851     bool copy_data;
852     const float* data;
853     const int *s_idx, *v_idx;
854 
855     if( !CV_IS_MAT(train_data) )
856         CV_ERROR( CV_StsBadArg, "Invalid or NULL training data matrix" );
857 
858     var_count = var_idx ? var_idx->cols + var_idx->rows - 1 :
859                 tflag == CV_ROW_SAMPLE ? train_data->cols : train_data->rows;
860     sample_count = sample_idx ? sample_idx->cols + sample_idx->rows - 1 :
861                    tflag == CV_ROW_SAMPLE ? train_data->rows : train_data->cols;
862 
863     if( _var_count )
864         *_var_count = var_count;
865 
866     if( _sample_count )
867         *_sample_count = sample_count;
868 
869     copy_data = tflag != CV_ROW_SAMPLE || var_idx || always_copy_data;
870 
871     CV_CALL( samples = (float**)cvAlloc(sample_count*sizeof(samples[0]) +
872                 (copy_data ? 1 : 0)*var_count*sample_count*sizeof(samples[0][0])) );
873     data = train_data->data.fl;
874     s_step = train_data->step / sizeof(samples[0][0]);
875     v_step = 1;
876     s_idx = sample_idx ? sample_idx->data.i : 0;
877     v_idx = var_idx ? var_idx->data.i : 0;
878 
879     if( !copy_data )
880     {
881         for( i = 0; i < sample_count; i++ )
882             samples[i] = (float*)(data + (s_idx ? s_idx[i] : i)*s_step);
883     }
884     else
885     {
886         samples[0] = (float*)(samples + sample_count);
887         if( tflag != CV_ROW_SAMPLE )
888             CV_SWAP( s_step, v_step, i );
889 
890         for( i = 0; i < sample_count; i++ )
891         {
892             float* dst = samples[i] = samples[0] + i*var_count;
893             const float* src = data + (s_idx ? s_idx[i] : i)*s_step;
894 
895             if( !v_idx )
896                 for( j = 0; j < var_count; j++ )
897                     dst[j] = src[j*v_step];
898             else
899                 for( j = 0; j < var_count; j++ )
900                     dst[j] = src[v_idx[j]*v_step];
901         }
902     }
903 
904     __END__;
905 
906     return (const float**)samples;
907 }
908 
909 
910 void
cvCheckTrainData(const CvMat * train_data,int tflag,const CvMat * missing_mask,int * var_all,int * sample_all)911 cvCheckTrainData( const CvMat* train_data, int tflag,
912                   const CvMat* missing_mask,
913                   int* var_all, int* sample_all )
914 {
915     CV_FUNCNAME( "cvCheckTrainData" );
916 
917     if( var_all )
918         *var_all = 0;
919 
920     if( sample_all )
921         *sample_all = 0;
922 
923     __BEGIN__;
924 
925     // check parameter types and sizes
926     if( !CV_IS_MAT(train_data) || CV_MAT_TYPE(train_data->type) != CV_32FC1 )
927         CV_ERROR( CV_StsBadArg, "train data must be floating-point matrix" );
928 
929     if( missing_mask )
930     {
931         if( !CV_IS_MAT(missing_mask) || !CV_IS_MASK_ARR(missing_mask) ||
932             !CV_ARE_SIZES_EQ(train_data, missing_mask) )
933             CV_ERROR( CV_StsBadArg,
934             "missing value mask must be 8-bit matrix of the same size as training data" );
935     }
936 
937     if( tflag != CV_ROW_SAMPLE && tflag != CV_COL_SAMPLE )
938         CV_ERROR( CV_StsBadArg,
939         "Unknown training data layout (must be CV_ROW_SAMPLE or CV_COL_SAMPLE)" );
940 
941     if( var_all )
942         *var_all = tflag == CV_ROW_SAMPLE ? train_data->cols : train_data->rows;
943 
944     if( sample_all )
945         *sample_all = tflag == CV_ROW_SAMPLE ? train_data->rows : train_data->cols;
946 
947     __END__;
948 }
949 
950 
951 int
cvPrepareTrainData(const char *,const CvMat * train_data,int tflag,const CvMat * responses,int response_type,const CvMat * var_idx,const CvMat * sample_idx,bool always_copy_data,const float *** out_train_samples,int * _sample_count,int * _var_count,int * _var_all,CvMat ** out_responses,CvMat ** out_response_map,CvMat ** out_var_idx,CvMat ** out_sample_idx)952 cvPrepareTrainData( const char* /*funcname*/,
953                     const CvMat* train_data, int tflag,
954                     const CvMat* responses, int response_type,
955                     const CvMat* var_idx,
956                     const CvMat* sample_idx,
957                     bool always_copy_data,
958                     const float*** out_train_samples,
959                     int* _sample_count,
960                     int* _var_count,
961                     int* _var_all,
962                     CvMat** out_responses,
963                     CvMat** out_response_map,
964                     CvMat** out_var_idx,
965                     CvMat** out_sample_idx )
966 {
967     int ok = 0;
968     CvMat* _var_idx = 0;
969     CvMat* _sample_idx = 0;
970     CvMat* _responses = 0;
971     int sample_all = 0, sample_count = 0, var_all = 0, var_count = 0;
972 
973     CV_FUNCNAME( "cvPrepareTrainData" );
974 
975     // step 0. clear all the output pointers to ensure we do not try
976     // to call free() with uninitialized pointers
977     if( out_responses )
978         *out_responses = 0;
979 
980     if( out_response_map )
981         *out_response_map = 0;
982 
983     if( out_var_idx )
984         *out_var_idx = 0;
985 
986     if( out_sample_idx )
987         *out_sample_idx = 0;
988 
989     if( out_train_samples )
990         *out_train_samples = 0;
991 
992     if( _sample_count )
993         *_sample_count = 0;
994 
995     if( _var_count )
996         *_var_count = 0;
997 
998     if( _var_all )
999         *_var_all = 0;
1000 
1001     __BEGIN__;
1002 
1003     if( !out_train_samples )
1004         CV_ERROR( CV_StsBadArg, "output pointer to train samples is NULL" );
1005 
1006     CV_CALL( cvCheckTrainData( train_data, tflag, 0, &var_all, &sample_all ));
1007 
1008     if( sample_idx )
1009         CV_CALL( _sample_idx = cvPreprocessIndexArray( sample_idx, sample_all ));
1010     if( var_idx )
1011         CV_CALL( _var_idx = cvPreprocessIndexArray( var_idx, var_all ));
1012 
1013     if( responses )
1014     {
1015         if( !out_responses )
1016             CV_ERROR( CV_StsNullPtr, "output response pointer is NULL" );
1017 
1018         if( response_type == CV_VAR_NUMERICAL )
1019         {
1020             CV_CALL( _responses = cvPreprocessOrderedResponses( responses,
1021                                                 _sample_idx, sample_all ));
1022         }
1023         else
1024         {
1025             CV_CALL( _responses = cvPreprocessCategoricalResponses( responses,
1026                                 _sample_idx, sample_all, out_response_map, 0 ));
1027         }
1028     }
1029 
1030     CV_CALL( *out_train_samples =
1031                 cvGetTrainSamples( train_data, tflag, _var_idx, _sample_idx,
1032                                    &var_count, &sample_count, always_copy_data ));
1033 
1034     ok = 1;
1035 
1036     __END__;
1037 
1038     if( ok )
1039     {
1040         if( out_responses )
1041             *out_responses = _responses, _responses = 0;
1042 
1043         if( out_var_idx )
1044             *out_var_idx = _var_idx, _var_idx = 0;
1045 
1046         if( out_sample_idx )
1047             *out_sample_idx = _sample_idx, _sample_idx = 0;
1048 
1049         if( _sample_count )
1050             *_sample_count = sample_count;
1051 
1052         if( _var_count )
1053             *_var_count = var_count;
1054 
1055         if( _var_all )
1056             *_var_all = var_all;
1057     }
1058     else
1059     {
1060         if( out_response_map )
1061             cvReleaseMat( out_response_map );
1062         cvFree( out_train_samples );
1063     }
1064 
1065     if( _responses != responses )
1066         cvReleaseMat( &_responses );
1067     cvReleaseMat( &_var_idx );
1068     cvReleaseMat( &_sample_idx );
1069 
1070     return ok;
1071 }
1072 
1073 
1074 typedef struct CvSampleResponsePair
1075 {
1076     const float* sample;
1077     const uchar* mask;
1078     int response;
1079     int index;
1080 }
1081 CvSampleResponsePair;
1082 
1083 
1084 static int
icvCmpSampleResponsePairs(const void * a,const void * b)1085 CV_CDECL icvCmpSampleResponsePairs( const void* a, const void* b )
1086 {
1087     int ra = ((const CvSampleResponsePair*)a)->response;
1088     int rb = ((const CvSampleResponsePair*)b)->response;
1089     int ia = ((const CvSampleResponsePair*)a)->index;
1090     int ib = ((const CvSampleResponsePair*)b)->index;
1091 
1092     return ra < rb ? -1 : ra > rb ? 1 : ia - ib;
1093     //return (ra > rb ? -1 : 0)|(ra < rb);
1094 }
1095 
1096 
1097 void
cvSortSamplesByClasses(const float ** samples,const CvMat * classes,int * class_ranges,const uchar ** mask)1098 cvSortSamplesByClasses( const float** samples, const CvMat* classes,
1099                         int* class_ranges, const uchar** mask )
1100 {
1101     CvSampleResponsePair* pairs = 0;
1102     CV_FUNCNAME( "cvSortSamplesByClasses" );
1103 
1104     __BEGIN__;
1105 
1106     int i, k = 0, sample_count;
1107 
1108     if( !samples || !classes || !class_ranges )
1109         CV_ERROR( CV_StsNullPtr, "INTERNAL ERROR: some of the args are NULL pointers" );
1110 
1111     if( classes->rows != 1 || CV_MAT_TYPE(classes->type) != CV_32SC1 )
1112         CV_ERROR( CV_StsBadArg, "classes array must be a single row of integers" );
1113 
1114     sample_count = classes->cols;
1115     CV_CALL( pairs = (CvSampleResponsePair*)cvAlloc( (sample_count+1)*sizeof(pairs[0])));
1116 
1117     for( i = 0; i < sample_count; i++ )
1118     {
1119         pairs[i].sample = samples[i];
1120         pairs[i].mask = (mask) ? (mask[i]) : 0;
1121         pairs[i].response = classes->data.i[i];
1122         pairs[i].index = i;
1123         assert( classes->data.i[i] >= 0 );
1124     }
1125 
1126     qsort( pairs, sample_count, sizeof(pairs[0]), icvCmpSampleResponsePairs );
1127     pairs[sample_count].response = -1;
1128     class_ranges[0] = 0;
1129 
1130     for( i = 0; i < sample_count; i++ )
1131     {
1132         samples[i] = pairs[i].sample;
1133         if (mask)
1134             mask[i] = pairs[i].mask;
1135         classes->data.i[i] = pairs[i].response;
1136 
1137         if( pairs[i].response != pairs[i+1].response )
1138             class_ranges[++k] = i+1;
1139     }
1140 
1141     __END__;
1142 
1143     cvFree( &pairs );
1144 }
1145 
1146 
1147 void
cvPreparePredictData(const CvArr * _sample,int dims_all,const CvMat * comp_idx,int class_count,const CvMat * prob,float ** _row_sample,int as_sparse)1148 cvPreparePredictData( const CvArr* _sample, int dims_all,
1149                       const CvMat* comp_idx, int class_count,
1150                       const CvMat* prob, float** _row_sample,
1151                       int as_sparse )
1152 {
1153     float* row_sample = 0;
1154     int* inverse_comp_idx = 0;
1155 
1156     CV_FUNCNAME( "cvPreparePredictData" );
1157 
1158     __BEGIN__;
1159 
1160     const CvMat* sample = (const CvMat*)_sample;
1161     float* sample_data;
1162     int sample_step;
1163     int is_sparse = CV_IS_SPARSE_MAT(sample);
1164     int d, sizes[CV_MAX_DIM];
1165     int i, dims_selected;
1166     int vec_size;
1167 
1168     if( !is_sparse && !CV_IS_MAT(sample) )
1169         CV_ERROR( !sample ? CV_StsNullPtr : CV_StsBadArg, "The sample is not a valid vector" );
1170 
1171     if( cvGetElemType( sample ) != CV_32FC1 )
1172         CV_ERROR( CV_StsUnsupportedFormat, "Input sample must have 32fC1 type" );
1173 
1174     CV_CALL( d = cvGetDims( sample, sizes ));
1175 
1176     if( !(is_sparse && d == 1 || !is_sparse && d == 2 && (sample->rows == 1 || sample->cols == 1)) )
1177         CV_ERROR( CV_StsBadSize, "Input sample must be 1-dimensional vector" );
1178 
1179     if( d == 1 )
1180         sizes[1] = 1;
1181 
1182     if( sizes[0] + sizes[1] - 1 != dims_all )
1183         CV_ERROR( CV_StsUnmatchedSizes,
1184         "The sample size is different from what has been used for training" );
1185 
1186     if( !_row_sample )
1187         CV_ERROR( CV_StsNullPtr, "INTERNAL ERROR: The row_sample pointer is NULL" );
1188 
1189     if( comp_idx && (!CV_IS_MAT(comp_idx) || comp_idx->rows != 1 ||
1190         CV_MAT_TYPE(comp_idx->type) != CV_32SC1) )
1191         CV_ERROR( CV_StsBadArg, "INTERNAL ERROR: invalid comp_idx" );
1192 
1193     dims_selected = comp_idx ? comp_idx->cols : dims_all;
1194 
1195     if( prob )
1196     {
1197         if( !CV_IS_MAT(prob) )
1198             CV_ERROR( CV_StsBadArg, "The output matrix of probabilities is invalid" );
1199 
1200         if( (prob->rows != 1 && prob->cols != 1) ||
1201             CV_MAT_TYPE(prob->type) != CV_32FC1 &&
1202             CV_MAT_TYPE(prob->type) != CV_64FC1 )
1203             CV_ERROR( CV_StsBadSize,
1204             "The matrix of probabilities must be 1-dimensional vector of 32fC1 type" );
1205 
1206         if( prob->rows + prob->cols - 1 != class_count )
1207             CV_ERROR( CV_StsUnmatchedSizes,
1208             "The vector of probabilities must contain as many elements as "
1209             "the number of classes in the training set" );
1210     }
1211 
1212     vec_size = !as_sparse ? dims_selected*sizeof(row_sample[0]) :
1213                 (dims_selected + 1)*sizeof(CvSparseVecElem32f);
1214 
1215     if( CV_IS_MAT(sample) )
1216     {
1217         sample_data = sample->data.fl;
1218         sample_step = sample->step / sizeof(row_sample[0]);
1219 
1220         if( !comp_idx && sample_step <= 1 && !as_sparse )
1221             *_row_sample = sample_data;
1222         else
1223         {
1224             CV_CALL( row_sample = (float*)cvAlloc( vec_size ));
1225 
1226             if( !comp_idx )
1227                 for( i = 0; i < dims_selected; i++ )
1228                     row_sample[i] = sample_data[sample_step*i];
1229             else
1230             {
1231                 int* comp = comp_idx->data.i;
1232                 if( !sample_step )
1233                     for( i = 0; i < dims_selected; i++ )
1234                         row_sample[i] = sample_data[comp[i]];
1235                 else
1236                     for( i = 0; i < dims_selected; i++ )
1237                         row_sample[i] = sample_data[sample_step*comp[i]];
1238             }
1239 
1240             *_row_sample = row_sample;
1241         }
1242 
1243         if( as_sparse )
1244         {
1245             const float* src = (const float*)row_sample;
1246             CvSparseVecElem32f* dst = (CvSparseVecElem32f*)row_sample;
1247 
1248             dst[dims_selected].idx = -1;
1249             for( i = dims_selected - 1; i >= 0; i-- )
1250             {
1251                 dst[i].idx = i;
1252                 dst[i].val = src[i];
1253             }
1254         }
1255     }
1256     else
1257     {
1258         CvSparseNode* node;
1259         CvSparseMatIterator mat_iterator;
1260         const CvSparseMat* sparse = (const CvSparseMat*)sample;
1261         assert( is_sparse );
1262 
1263         node = cvInitSparseMatIterator( sparse, &mat_iterator );
1264         CV_CALL( row_sample = (float*)cvAlloc( vec_size ));
1265 
1266         if( comp_idx )
1267         {
1268             CV_CALL( inverse_comp_idx = (int*)cvAlloc( dims_all*sizeof(int) ));
1269             memset( inverse_comp_idx, -1, dims_all*sizeof(int) );
1270             for( i = 0; i < dims_selected; i++ )
1271                 inverse_comp_idx[comp_idx->data.i[i]] = i;
1272         }
1273 
1274         if( !as_sparse )
1275         {
1276             memset( row_sample, 0, vec_size );
1277 
1278             for( ; node != 0; node = cvGetNextSparseNode(&mat_iterator) )
1279             {
1280                 int idx = *CV_NODE_IDX( sparse, node );
1281                 if( inverse_comp_idx )
1282                 {
1283                     idx = inverse_comp_idx[idx];
1284                     if( idx < 0 )
1285                         continue;
1286                 }
1287                 row_sample[idx] = *(float*)CV_NODE_VAL( sparse, node );
1288             }
1289         }
1290         else
1291         {
1292             CvSparseVecElem32f* ptr = (CvSparseVecElem32f*)row_sample;
1293 
1294             for( ; node != 0; node = cvGetNextSparseNode(&mat_iterator) )
1295             {
1296                 int idx = *CV_NODE_IDX( sparse, node );
1297                 if( inverse_comp_idx )
1298                 {
1299                     idx = inverse_comp_idx[idx];
1300                     if( idx < 0 )
1301                         continue;
1302                 }
1303                 ptr->idx = idx;
1304                 ptr->val = *(float*)CV_NODE_VAL( sparse, node );
1305                 ptr++;
1306             }
1307 
1308             qsort( row_sample, ptr - (CvSparseVecElem32f*)row_sample,
1309                    sizeof(ptr[0]), icvCmpSparseVecElems );
1310             ptr->idx = -1;
1311         }
1312 
1313         *_row_sample = row_sample;
1314     }
1315 
1316     __END__;
1317 
1318     if( inverse_comp_idx )
1319         cvFree( &inverse_comp_idx );
1320 
1321     if( cvGetErrStatus() < 0 && _row_sample )
1322     {
1323         cvFree( &row_sample );
1324         *_row_sample = 0;
1325     }
1326 }
1327 
1328 
1329 static void
icvConvertDataToSparse(const uchar * src,int src_step,int src_type,uchar * dst,int dst_step,int dst_type,CvSize size,int * idx)1330 icvConvertDataToSparse( const uchar* src, int src_step, int src_type,
1331                         uchar* dst, int dst_step, int dst_type,
1332                         CvSize size, int* idx )
1333 {
1334     CV_FUNCNAME( "icvConvertDataToSparse" );
1335 
1336     __BEGIN__;
1337 
1338     int i, j;
1339     src_type = CV_MAT_TYPE(src_type);
1340     dst_type = CV_MAT_TYPE(dst_type);
1341 
1342     if( CV_MAT_CN(src_type) != 1 || CV_MAT_CN(dst_type) != 1 )
1343         CV_ERROR( CV_StsUnsupportedFormat, "The function supports only single-channel arrays" );
1344 
1345     if( src_step == 0 )
1346         src_step = CV_ELEM_SIZE(src_type);
1347 
1348     if( dst_step == 0 )
1349         dst_step = CV_ELEM_SIZE(dst_type);
1350 
1351     // if there is no "idx" and if both arrays are continuous,
1352     // do the whole processing (copying or conversion) in a single loop
1353     if( !idx && CV_ELEM_SIZE(src_type)*size.width == src_step &&
1354         CV_ELEM_SIZE(dst_type)*size.width == dst_step )
1355     {
1356         size.width *= size.height;
1357         size.height = 1;
1358     }
1359 
1360     if( src_type == dst_type )
1361     {
1362         int full_width = CV_ELEM_SIZE(dst_type)*size.width;
1363 
1364         if( full_width == sizeof(int) ) // another common case: copy int's or float's
1365             for( i = 0; i < size.height; i++, src += src_step )
1366                 *(int*)(dst + dst_step*(idx ? idx[i] : i)) = *(int*)src;
1367         else
1368             for( i = 0; i < size.height; i++, src += src_step )
1369                 memcpy( dst + dst_step*(idx ? idx[i] : i), src, full_width );
1370     }
1371     else if( src_type == CV_32SC1 && (dst_type == CV_32FC1 || dst_type == CV_64FC1) )
1372         for( i = 0; i < size.height; i++, src += src_step )
1373         {
1374             uchar* _dst = dst + dst_step*(idx ? idx[i] : i);
1375             if( dst_type == CV_32FC1 )
1376                 for( j = 0; j < size.width; j++ )
1377                     ((float*)_dst)[j] = (float)((int*)src)[j];
1378             else
1379                 for( j = 0; j < size.width; j++ )
1380                     ((double*)_dst)[j] = ((int*)src)[j];
1381         }
1382     else if( (src_type == CV_32FC1 || src_type == CV_64FC1) && dst_type == CV_32SC1 )
1383         for( i = 0; i < size.height; i++, src += src_step )
1384         {
1385             uchar* _dst = dst + dst_step*(idx ? idx[i] : i);
1386             if( src_type == CV_32FC1 )
1387                 for( j = 0; j < size.width; j++ )
1388                     ((int*)_dst)[j] = cvRound(((float*)src)[j]);
1389             else
1390                 for( j = 0; j < size.width; j++ )
1391                     ((int*)_dst)[j] = cvRound(((double*)src)[j]);
1392         }
1393     else if( src_type == CV_32FC1 && dst_type == CV_64FC1 ||
1394              src_type == CV_64FC1 && dst_type == CV_32FC1 )
1395         for( i = 0; i < size.height; i++, src += src_step )
1396         {
1397             uchar* _dst = dst + dst_step*(idx ? idx[i] : i);
1398             if( src_type == CV_32FC1 )
1399                 for( j = 0; j < size.width; j++ )
1400                     ((double*)_dst)[j] = ((float*)src)[j];
1401             else
1402                 for( j = 0; j < size.width; j++ )
1403                     ((float*)_dst)[j] = (float)((double*)src)[j];
1404         }
1405     else
1406         CV_ERROR( CV_StsUnsupportedFormat, "Unsupported combination of input and output vectors" );
1407 
1408     __END__;
1409 }
1410 
1411 
1412 void
cvWritebackLabels(const CvMat * labels,CvMat * dst_labels,const CvMat * centers,CvMat * dst_centers,const CvMat * probs,CvMat * dst_probs,const CvMat * sample_idx,int samples_all,const CvMat * comp_idx,int dims_all)1413 cvWritebackLabels( const CvMat* labels, CvMat* dst_labels,
1414                    const CvMat* centers, CvMat* dst_centers,
1415                    const CvMat* probs, CvMat* dst_probs,
1416                    const CvMat* sample_idx, int samples_all,
1417                    const CvMat* comp_idx, int dims_all )
1418 {
1419     CV_FUNCNAME( "cvWritebackLabels" );
1420 
1421     __BEGIN__;
1422 
1423     int samples_selected = samples_all, dims_selected = dims_all;
1424 
1425     if( dst_labels && !CV_IS_MAT(dst_labels) )
1426         CV_ERROR( CV_StsBadArg, "Array of output labels is not a valid matrix" );
1427 
1428     if( dst_centers )
1429         if( !ICV_IS_MAT_OF_TYPE(dst_centers, CV_32FC1) &&
1430             !ICV_IS_MAT_OF_TYPE(dst_centers, CV_64FC1) )
1431             CV_ERROR( CV_StsBadArg, "Array of cluster centers is not a valid matrix" );
1432 
1433     if( dst_probs && !CV_IS_MAT(dst_probs) )
1434         CV_ERROR( CV_StsBadArg, "Probability matrix is not valid" );
1435 
1436     if( sample_idx )
1437     {
1438         CV_ASSERT( sample_idx->rows == 1 && CV_MAT_TYPE(sample_idx->type) == CV_32SC1 );
1439         samples_selected = sample_idx->cols;
1440     }
1441 
1442     if( comp_idx )
1443     {
1444         CV_ASSERT( comp_idx->rows == 1 && CV_MAT_TYPE(comp_idx->type) == CV_32SC1 );
1445         dims_selected = comp_idx->cols;
1446     }
1447 
1448     if( dst_labels && (!labels || labels->data.ptr != dst_labels->data.ptr) )
1449     {
1450         if( !labels )
1451             CV_ERROR( CV_StsNullPtr, "NULL labels" );
1452 
1453         CV_ASSERT( labels->rows == 1 );
1454 
1455         if( dst_labels->rows != 1 && dst_labels->cols != 1 )
1456             CV_ERROR( CV_StsBadSize, "Array of output labels should be 1d vector" );
1457 
1458         if( dst_labels->rows + dst_labels->cols - 1 != samples_all )
1459             CV_ERROR( CV_StsUnmatchedSizes,
1460             "Size of vector of output labels is not equal to the total number of input samples" );
1461 
1462         CV_ASSERT( labels->cols == samples_selected );
1463 
1464         CV_CALL( icvConvertDataToSparse( labels->data.ptr, labels->step, labels->type,
1465                         dst_labels->data.ptr, dst_labels->step, dst_labels->type,
1466                         cvSize( 1, samples_selected ), sample_idx ? sample_idx->data.i : 0 ));
1467     }
1468 
1469     if( dst_centers && (!centers || centers->data.ptr != dst_centers->data.ptr) )
1470     {
1471         int i;
1472 
1473         if( !centers )
1474             CV_ERROR( CV_StsNullPtr, "NULL centers" );
1475 
1476         if( centers->rows != dst_centers->rows )
1477             CV_ERROR( CV_StsUnmatchedSizes, "Invalid number of rows in matrix of output centers" );
1478 
1479         if( dst_centers->cols != dims_all )
1480             CV_ERROR( CV_StsUnmatchedSizes,
1481             "Number of columns in matrix of output centers is "
1482             "not equal to the total number of components in the input samples" );
1483 
1484         CV_ASSERT( centers->cols == dims_selected );
1485 
1486         for( i = 0; i < centers->rows; i++ )
1487             CV_CALL( icvConvertDataToSparse( centers->data.ptr + i*centers->step, 0, centers->type,
1488                         dst_centers->data.ptr + i*dst_centers->step, 0, dst_centers->type,
1489                         cvSize( 1, dims_selected ), comp_idx ? comp_idx->data.i : 0 ));
1490     }
1491 
1492     if( dst_probs && (!probs || probs->data.ptr != dst_probs->data.ptr) )
1493     {
1494         if( !probs )
1495             CV_ERROR( CV_StsNullPtr, "NULL probs" );
1496 
1497         if( probs->cols != dst_probs->cols )
1498             CV_ERROR( CV_StsUnmatchedSizes, "Invalid number of columns in output probability matrix" );
1499 
1500         if( dst_probs->rows != samples_all )
1501             CV_ERROR( CV_StsUnmatchedSizes,
1502             "Number of rows in output probability matrix is "
1503             "not equal to the total number of input samples" );
1504 
1505         CV_ASSERT( probs->rows == samples_selected );
1506 
1507         CV_CALL( icvConvertDataToSparse( probs->data.ptr, probs->step, probs->type,
1508                         dst_probs->data.ptr, dst_probs->step, dst_probs->type,
1509                         cvSize( probs->cols, samples_selected ),
1510                         sample_idx ? sample_idx->data.i : 0 ));
1511     }
1512 
1513     __END__;
1514 }
1515 
1516 #if 0
1517 CV_IMPL void
1518 cvStatModelMultiPredict( const CvStatModel* stat_model,
1519                          const CvArr* predict_input,
1520                          int flags, CvMat* predict_output,
1521                          CvMat* probs, const CvMat* sample_idx )
1522 {
1523     CvMemStorage* storage = 0;
1524     CvMat* sample_idx_buffer = 0;
1525     CvSparseMat** sparse_rows = 0;
1526     int samples_selected = 0;
1527 
1528     CV_FUNCNAME( "cvStatModelMultiPredict" );
1529 
1530     __BEGIN__;
1531 
1532     int i;
1533     int predict_output_step = 1, sample_idx_step = 1;
1534     int type;
1535     int d, sizes[CV_MAX_DIM];
1536     int tflag = flags == CV_COL_SAMPLE;
1537     int samples_all, dims_all;
1538     int is_sparse = CV_IS_SPARSE_MAT(predict_input);
1539     CvMat predict_input_part;
1540     CvArr* sample = &predict_input_part;
1541     CvMat probs_part;
1542     CvMat* probs1 = probs ? &probs_part : 0;
1543 
1544     if( !CV_IS_STAT_MODEL(stat_model) )
1545         CV_ERROR( !stat_model ? CV_StsNullPtr : CV_StsBadArg, "Invalid statistical model" );
1546 
1547     if( !stat_model->predict )
1548         CV_ERROR( CV_StsNotImplemented, "There is no \"predict\" method" );
1549 
1550     if( !predict_input || !predict_output )
1551         CV_ERROR( CV_StsNullPtr, "NULL input or output matrices" );
1552 
1553     if( !is_sparse && !CV_IS_MAT(predict_input) )
1554         CV_ERROR( CV_StsBadArg, "predict_input should be a matrix or a sparse matrix" );
1555 
1556     if( !CV_IS_MAT(predict_output) )
1557         CV_ERROR( CV_StsBadArg, "predict_output should be a matrix" );
1558 
1559     type = cvGetElemType( predict_input );
1560     if( type != CV_32FC1 ||
1561         (CV_MAT_TYPE(predict_output->type) != CV_32FC1 &&
1562          CV_MAT_TYPE(predict_output->type) != CV_32SC1 ))
1563          CV_ERROR( CV_StsUnsupportedFormat, "The input or output matrix has unsupported format" );
1564 
1565     CV_CALL( d = cvGetDims( predict_input, sizes ));
1566     if( d > 2 )
1567         CV_ERROR( CV_StsBadSize, "The input matrix should be 1- or 2-dimensional" );
1568 
1569     if( !tflag )
1570     {
1571         samples_all = samples_selected = sizes[0];
1572         dims_all = sizes[1];
1573     }
1574     else
1575     {
1576         samples_all = samples_selected = sizes[1];
1577         dims_all = sizes[0];
1578     }
1579 
1580     if( sample_idx )
1581     {
1582         if( !CV_IS_MAT(sample_idx) )
1583             CV_ERROR( CV_StsBadArg, "Invalid sample_idx matrix" );
1584 
1585         if( sample_idx->cols != 1 && sample_idx->rows != 1 )
1586             CV_ERROR( CV_StsBadSize, "sample_idx must be 1-dimensional matrix" );
1587 
1588         samples_selected = sample_idx->rows + sample_idx->cols - 1;
1589 
1590         if( CV_MAT_TYPE(sample_idx->type) == CV_32SC1 )
1591         {
1592             if( samples_selected > samples_all )
1593                 CV_ERROR( CV_StsBadSize, "sample_idx is too large vector" );
1594         }
1595         else if( samples_selected != samples_all )
1596             CV_ERROR( CV_StsUnmatchedSizes, "sample_idx has incorrect size" );
1597 
1598         sample_idx_step = sample_idx->step ?
1599             sample_idx->step / CV_ELEM_SIZE(sample_idx->type) : 1;
1600     }
1601 
1602     if( predict_output->rows != 1 && predict_output->cols != 1 )
1603         CV_ERROR( CV_StsBadSize, "predict_output should be a 1-dimensional matrix" );
1604 
1605     if( predict_output->rows + predict_output->cols - 1 != samples_all )
1606         CV_ERROR( CV_StsUnmatchedSizes, "predict_output and predict_input have uncoordinated sizes" );
1607 
1608     predict_output_step = predict_output->step ?
1609         predict_output->step / CV_ELEM_SIZE(predict_output->type) : 1;
1610 
1611     if( probs )
1612     {
1613         if( !CV_IS_MAT(probs) )
1614             CV_ERROR( CV_StsBadArg, "Invalid matrix of probabilities" );
1615 
1616         if( probs->rows != samples_all )
1617             CV_ERROR( CV_StsUnmatchedSizes,
1618             "matrix of probabilities must have as many rows as the total number of samples" );
1619 
1620         if( CV_MAT_TYPE(probs->type) != CV_32FC1 )
1621             CV_ERROR( CV_StsUnsupportedFormat, "matrix of probabilities must have 32fC1 type" );
1622     }
1623 
1624     if( is_sparse )
1625     {
1626         CvSparseNode* node;
1627         CvSparseMatIterator mat_iterator;
1628         CvSparseMat* sparse = (CvSparseMat*)predict_input;
1629 
1630         if( sample_idx && CV_MAT_TYPE(sample_idx->type) == CV_32SC1 )
1631         {
1632             CV_CALL( sample_idx_buffer = cvCreateMat( 1, samples_all, CV_8UC1 ));
1633             cvZero( sample_idx_buffer );
1634             for( i = 0; i < samples_selected; i++ )
1635                 sample_idx_buffer->data.ptr[sample_idx->data.i[i*sample_idx_step]] = 1;
1636             samples_selected = samples_all;
1637             sample_idx = sample_idx_buffer;
1638             sample_idx_step = 1;
1639         }
1640 
1641         CV_CALL( sparse_rows = (CvSparseMat**)cvAlloc( samples_selected*sizeof(sparse_rows[0])));
1642         for( i = 0; i < samples_selected; i++ )
1643         {
1644             if( sample_idx && sample_idx->data.ptr[i*sample_idx_step] == 0 )
1645                 continue;
1646             CV_CALL( sparse_rows[i] = cvCreateSparseMat( 1, &dims_all, type ));
1647             if( !storage )
1648                 storage = sparse_rows[i]->heap->storage;
1649             else
1650             {
1651                 // hack: to decrease memory footprint, make all the sparse matrices
1652                 // reside in the same storage
1653                 int elem_size = sparse_rows[i]->heap->elem_size;
1654                 cvReleaseMemStorage( &sparse_rows[i]->heap->storage );
1655                 sparse_rows[i]->heap = cvCreateSet( 0, sizeof(CvSet), elem_size, storage );
1656             }
1657         }
1658 
1659         // put each row (or column) of predict_input into separate sparse matrix.
1660         node = cvInitSparseMatIterator( sparse, &mat_iterator );
1661         for( ; node != 0; node = cvGetNextSparseNode( &mat_iterator ))
1662         {
1663             int* idx = CV_NODE_IDX( sparse, node );
1664             int idx0 = idx[tflag ^ 1];
1665             int idx1 = idx[tflag];
1666 
1667             if( sample_idx && sample_idx->data.ptr[idx0*sample_idx_step] == 0 )
1668                 continue;
1669 
1670             assert( sparse_rows[idx0] != 0 );
1671             *(float*)cvPtrND( sparse, &idx1, 0, 1, 0 ) = *(float*)CV_NODE_VAL( sparse, node );
1672         }
1673     }
1674 
1675     for( i = 0; i < samples_selected; i++ )
1676     {
1677         int idx = i;
1678         float response;
1679 
1680         if( sample_idx )
1681         {
1682             if( CV_MAT_TYPE(sample_idx->type) == CV_32SC1 )
1683             {
1684                 idx = sample_idx->data.i[i*sample_idx_step];
1685                 if( (unsigned)idx >= (unsigned)samples_all )
1686                     CV_ERROR( CV_StsOutOfRange, "Some of sample_idx elements are out of range" );
1687             }
1688             else if( CV_MAT_TYPE(sample_idx->type) == CV_8UC1 &&
1689                      sample_idx->data.ptr[i*sample_idx_step] == 0 )
1690                 continue;
1691         }
1692 
1693         if( !is_sparse )
1694         {
1695             if( !tflag )
1696                 cvGetRow( predict_input, &predict_input_part, idx );
1697             else
1698             {
1699                 cvGetCol( predict_input, &predict_input_part, idx );
1700             }
1701         }
1702         else
1703             sample = sparse_rows[idx];
1704 
1705         if( probs )
1706             cvGetRow( probs, probs1, idx );
1707 
1708         CV_CALL( response = stat_model->predict( stat_model, (CvMat*)sample, probs1 ));
1709 
1710         if( CV_MAT_TYPE(predict_output->type) == CV_32FC1 )
1711             predict_output->data.fl[idx*predict_output_step] = response;
1712         else
1713         {
1714             CV_ASSERT( cvRound(response) == response );
1715             predict_output->data.i[idx*predict_output_step] = cvRound(response);
1716         }
1717     }
1718 
1719     __END__;
1720 
1721     if( sparse_rows )
1722     {
1723         int i;
1724         for( i = 0; i < samples_selected; i++ )
1725             if( sparse_rows[i] )
1726             {
1727                 sparse_rows[i]->heap->storage = 0;
1728                 cvReleaseSparseMat( &sparse_rows[i] );
1729             }
1730         cvFree( &sparse_rows );
1731     }
1732 
1733     cvReleaseMat( &sample_idx_buffer );
1734     cvReleaseMemStorage( &storage );
1735 }
1736 #endif
1737 
1738 // By P. Yarykin - begin -
1739 
cvCombineResponseMaps(CvMat * _responses,const CvMat * old_response_map,CvMat * new_response_map,CvMat ** out_response_map)1740 void cvCombineResponseMaps (CvMat*  _responses,
1741                       const CvMat*  old_response_map,
1742                             CvMat*  new_response_map,
1743                             CvMat** out_response_map)
1744 {
1745     int** old_data = NULL;
1746     int** new_data = NULL;
1747 
1748         CV_FUNCNAME ("cvCombineResponseMaps");
1749         __BEGIN__
1750 
1751     int i,j;
1752     int old_n, new_n, out_n;
1753     int samples, free_response;
1754     int* first;
1755     int* responses;
1756     int* out_data;
1757 
1758     if( out_response_map )
1759         *out_response_map = 0;
1760 
1761 // Check input data.
1762     if ((!ICV_IS_MAT_OF_TYPE (_responses, CV_32SC1)) ||
1763         (!ICV_IS_MAT_OF_TYPE (old_response_map, CV_32SC1)) ||
1764         (!ICV_IS_MAT_OF_TYPE (new_response_map, CV_32SC1)))
1765     {
1766         CV_ERROR (CV_StsBadArg, "Some of input arguments is not the CvMat")
1767     }
1768 
1769 // Prepare sorted responses.
1770     first = new_response_map->data.i;
1771     new_n = new_response_map->cols;
1772     CV_CALL (new_data = (int**)cvAlloc (new_n * sizeof (new_data[0])));
1773     for (i = 0; i < new_n; i++)
1774         new_data[i] = first + i;
1775     qsort (new_data, new_n, sizeof(int*), icvCmpIntegersPtr);
1776 
1777     first = old_response_map->data.i;
1778     old_n = old_response_map->cols;
1779     CV_CALL (old_data = (int**)cvAlloc (old_n * sizeof (old_data[0])));
1780     for (i = 0; i < old_n; i++)
1781         old_data[i] = first + i;
1782     qsort (old_data, old_n, sizeof(int*), icvCmpIntegersPtr);
1783 
1784 // Count the number of different responses.
1785     for (i = 0, j = 0, out_n = 0; i < old_n && j < new_n; out_n++)
1786     {
1787         if (*old_data[i] == *new_data[j])
1788         {
1789             i++;
1790             j++;
1791         }
1792         else if (*old_data[i] < *new_data[j])
1793             i++;
1794         else
1795             j++;
1796     }
1797     out_n += old_n - i + new_n - j;
1798 
1799 // Create and fill the result response maps.
1800     CV_CALL (*out_response_map = cvCreateMat (1, out_n, CV_32SC1));
1801     out_data = (*out_response_map)->data.i;
1802     memcpy (out_data, first, old_n * sizeof (int));
1803 
1804     free_response = old_n;
1805     for (i = 0, j = 0; i < old_n && j < new_n; )
1806     {
1807         if (*old_data[i] == *new_data[j])
1808         {
1809             *new_data[j] = (int)(old_data[i] - first);
1810             i++;
1811             j++;
1812         }
1813         else if (*old_data[i] < *new_data[j])
1814             i++;
1815         else
1816         {
1817             out_data[free_response] = *new_data[j];
1818             *new_data[j] = free_response++;
1819             j++;
1820         }
1821     }
1822     for (; j < new_n; j++)
1823     {
1824         out_data[free_response] = *new_data[j];
1825         *new_data[j] = free_response++;
1826     }
1827     CV_ASSERT (free_response == out_n);
1828 
1829 // Change <responses> according to out response map.
1830     samples = _responses->cols + _responses->rows - 1;
1831     responses = _responses->data.i;
1832     first = new_response_map->data.i;
1833     for (i = 0; i < samples; i++)
1834     {
1835         responses[i] = first[responses[i]];
1836     }
1837 
1838         __END__
1839 
1840     cvFree(&old_data);
1841     cvFree(&new_data);
1842 
1843 }
1844 
1845 
icvGetNumberOfCluster(double * prob_vector,int num_of_clusters,float r,float outlier_thresh,int normalize_probs)1846 int icvGetNumberOfCluster( double* prob_vector, int num_of_clusters, float r,
1847                            float outlier_thresh, int normalize_probs )
1848 {
1849     int max_prob_loc = 0;
1850 
1851     CV_FUNCNAME("icvGetNumberOfCluster");
1852     __BEGIN__;
1853 
1854     double prob, maxprob, sum;
1855     int i;
1856 
1857     CV_ASSERT(prob_vector);
1858     CV_ASSERT(num_of_clusters >= 0);
1859 
1860     maxprob = prob_vector[0];
1861     max_prob_loc = 0;
1862     sum = maxprob;
1863     for( i = 1; i < num_of_clusters; i++ )
1864     {
1865         prob = prob_vector[i];
1866         sum += prob;
1867         if( prob > maxprob )
1868         {
1869             max_prob_loc = i;
1870             maxprob = prob;
1871         }
1872     }
1873     if( normalize_probs && fabs(sum - 1.) > FLT_EPSILON )
1874     {
1875         for( i = 0; i < num_of_clusters; i++ )
1876             prob_vector[i] /= sum;
1877     }
1878     if( fabs(r - 1.) > FLT_EPSILON && fabs(sum - 1.) < outlier_thresh )
1879         max_prob_loc = -1;
1880 
1881     __END__;
1882 
1883     return max_prob_loc;
1884 
1885 } // End of icvGetNumberOfCluster
1886 
1887 
icvFindClusterLabels(const CvMat * probs,float outlier_thresh,float r,const CvMat * labels)1888 void icvFindClusterLabels( const CvMat* probs, float outlier_thresh, float r,
1889                           const CvMat* labels )
1890 {
1891     CvMat* counts = 0;
1892 
1893     CV_FUNCNAME("icvFindClusterLabels");
1894     __BEGIN__;
1895 
1896     int nclusters, nsamples;
1897     int i, j;
1898     double* probs_data;
1899 
1900     CV_ASSERT( ICV_IS_MAT_OF_TYPE(probs, CV_64FC1) );
1901     CV_ASSERT( ICV_IS_MAT_OF_TYPE(labels, CV_32SC1) );
1902 
1903     nclusters = probs->cols;
1904     nsamples  = probs->rows;
1905     CV_ASSERT( nsamples == labels->cols );
1906 
1907     CV_CALL( counts = cvCreateMat( 1, nclusters + 1, CV_32SC1 ) );
1908     CV_CALL( cvSetZero( counts ));
1909     for( i = 0; i < nsamples; i++ )
1910     {
1911         labels->data.i[i] = icvGetNumberOfCluster( probs->data.db + i*probs->cols,
1912             nclusters, r, outlier_thresh, 1 );
1913         counts->data.i[labels->data.i[i] + 1]++;
1914     }
1915     CV_ASSERT((int)cvSum(counts).val[0] == nsamples);
1916     // Filling empty clusters with the vector, that has the maximal probability
1917     for( j = 0; j < nclusters; j++ ) // outliers are ignored
1918     {
1919         int maxprob_loc = -1;
1920         double maxprob = 0;
1921 
1922         if( counts->data.i[j+1] ) // j-th class is not empty
1923             continue;
1924         // look for the presentative, which is not lonely in it's cluster
1925         // and that has a maximal probability among all these vectors
1926         probs_data = probs->data.db;
1927         for( i = 0; i < nsamples; i++, probs_data++ )
1928         {
1929             int label = labels->data.i[i];
1930             double prob;
1931             if( counts->data.i[label+1] == 0 ||
1932                 (counts->data.i[label+1] <= 1 && label != -1) )
1933                 continue;
1934             prob = *probs_data;
1935             if( prob >= maxprob )
1936             {
1937                 maxprob = prob;
1938                 maxprob_loc = i;
1939             }
1940         }
1941         // maxprob_loc == 0 <=> number of vectors less then number of clusters
1942         CV_ASSERT( maxprob_loc >= 0 );
1943         counts->data.i[labels->data.i[maxprob_loc] + 1]--;
1944         labels->data.i[maxprob_loc] = j;
1945         counts->data.i[j + 1]++;
1946     }
1947 
1948     __END__;
1949 
1950     cvReleaseMat( &counts );
1951 } // End of icvFindClusterLabels
1952 
1953 /* End of file */
1954