• 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 //                For Open Source Computer Vision Library
12 //
13 // Copyright (C) 2000, Intel Corporation, all rights reserved.
14 // Third party copyrights are property of their respective owners.
15 //
16 // Redistribution and use in source and binary forms, with or without modification,
17 // are permitted provided that the following conditions are met:
18 //
19 //   * Redistribution's of source code must retain the above copyright notice,
20 //     this list of conditions and the following disclaimer.
21 //
22 //   * Redistribution's in binary form must reproduce the above copyright notice,
23 //     this list of conditions and the following disclaimer in the documentation
24 //     and/or other materials provided with the distribution.
25 //
26 //   * The name of Intel Corporation may not be used to endorse or promote products
27 //     derived from this software without specific prior written permission.
28 //
29 // This software is provided by the copyright holders and contributors "as is" and
30 // any express or implied warranties, including, but not limited to, the implied
31 // warranties of merchantability and fitness for a particular purpose are disclaimed.
32 // In no event shall the Intel Corporation or contributors be liable for any direct,
33 // indirect, incidental, special, exemplary, or consequential damages
34 // (including, but not limited to, procurement of substitute goods or services;
35 // loss of use, data, or profits; or business interruption) however caused
36 // and on any theory of liability, whether in contract, strict liability,
37 // or tort (including negligence or otherwise) arising in any way out of
38 // the use of this software, even if advised of the possibility of such damage.
39 //
40 //M*/
41 
42 /* Haar features calculation */
43 
44 #include "_cv.h"
45 #include <stdio.h>
46 
47 /* these settings affect the quality of detection: change with care */
48 #define CV_ADJUST_FEATURES 1
49 #define CV_ADJUST_WEIGHTS  0
50 
51 typedef int sumtype;
52 typedef double sqsumtype;
53 
54 typedef struct CvHidHaarFeature
55 {
56     struct
57     {
58         sumtype *p0, *p1, *p2, *p3;
59         float weight;
60     }
61     rect[CV_HAAR_FEATURE_MAX];
62 }
63 CvHidHaarFeature;
64 
65 
66 typedef struct CvHidHaarTreeNode
67 {
68     CvHidHaarFeature feature;
69     float threshold;
70     int left;
71     int right;
72 }
73 CvHidHaarTreeNode;
74 
75 
76 typedef struct CvHidHaarClassifier
77 {
78     int count;
79     //CvHaarFeature* orig_feature;
80     CvHidHaarTreeNode* node;
81     float* alpha;
82 }
83 CvHidHaarClassifier;
84 
85 
86 typedef struct CvHidHaarStageClassifier
87 {
88     int  count;
89     float threshold;
90     CvHidHaarClassifier* classifier;
91     int two_rects;
92 
93     struct CvHidHaarStageClassifier* next;
94     struct CvHidHaarStageClassifier* child;
95     struct CvHidHaarStageClassifier* parent;
96 }
97 CvHidHaarStageClassifier;
98 
99 
100 struct CvHidHaarClassifierCascade
101 {
102     int  count;
103     int  is_stump_based;
104     int  has_tilted_features;
105     int  is_tree;
106     double inv_window_area;
107     CvMat sum, sqsum, tilted;
108     CvHidHaarStageClassifier* stage_classifier;
109     sqsumtype *pq0, *pq1, *pq2, *pq3;
110     sumtype *p0, *p1, *p2, *p3;
111 
112     void** ipp_stages;
113 };
114 
115 
116 /* IPP functions for object detection */
117 icvHaarClassifierInitAlloc_32f_t icvHaarClassifierInitAlloc_32f_p = 0;
118 icvHaarClassifierFree_32f_t icvHaarClassifierFree_32f_p = 0;
119 icvApplyHaarClassifier_32f_C1R_t icvApplyHaarClassifier_32f_C1R_p = 0;
120 icvRectStdDev_32f_C1R_t icvRectStdDev_32f_C1R_p = 0;
121 
122 const int icv_object_win_border = 1;
123 const float icv_stage_threshold_bias = 0.0001f;
124 
125 static CvHaarClassifierCascade*
icvCreateHaarClassifierCascade(int stage_count)126 icvCreateHaarClassifierCascade( int stage_count )
127 {
128     CvHaarClassifierCascade* cascade = 0;
129 
130     CV_FUNCNAME( "icvCreateHaarClassifierCascade" );
131 
132     __BEGIN__;
133 
134     int block_size = sizeof(*cascade) + stage_count*sizeof(*cascade->stage_classifier);
135 
136     if( stage_count <= 0 )
137         CV_ERROR( CV_StsOutOfRange, "Number of stages should be positive" );
138 
139     CV_CALL( cascade = (CvHaarClassifierCascade*)cvAlloc( block_size ));
140     memset( cascade, 0, block_size );
141 
142     cascade->stage_classifier = (CvHaarStageClassifier*)(cascade + 1);
143     cascade->flags = CV_HAAR_MAGIC_VAL;
144     cascade->count = stage_count;
145 
146     __END__;
147 
148     return cascade;
149 }
150 
151 static void
icvReleaseHidHaarClassifierCascade(CvHidHaarClassifierCascade ** _cascade)152 icvReleaseHidHaarClassifierCascade( CvHidHaarClassifierCascade** _cascade )
153 {
154     if( _cascade && *_cascade )
155     {
156         CvHidHaarClassifierCascade* cascade = *_cascade;
157         if( cascade->ipp_stages && icvHaarClassifierFree_32f_p )
158         {
159             int i;
160             for( i = 0; i < cascade->count; i++ )
161             {
162                 if( cascade->ipp_stages[i] )
163                     icvHaarClassifierFree_32f_p( cascade->ipp_stages[i] );
164             }
165         }
166         cvFree( &cascade->ipp_stages );
167         cvFree( _cascade );
168     }
169 }
170 
171 /* create more efficient internal representation of haar classifier cascade */
172 static CvHidHaarClassifierCascade*
icvCreateHidHaarClassifierCascade(CvHaarClassifierCascade * cascade)173 icvCreateHidHaarClassifierCascade( CvHaarClassifierCascade* cascade )
174 {
175     CvRect* ipp_features = 0;
176     float *ipp_weights = 0, *ipp_thresholds = 0, *ipp_val1 = 0, *ipp_val2 = 0;
177     int* ipp_counts = 0;
178 
179     CvHidHaarClassifierCascade* out = 0;
180 
181     CV_FUNCNAME( "icvCreateHidHaarClassifierCascade" );
182 
183     __BEGIN__;
184 
185     int i, j, k, l;
186     int datasize;
187     int total_classifiers = 0;
188     int total_nodes = 0;
189     char errorstr[100];
190     CvHidHaarClassifier* haar_classifier_ptr;
191     CvHidHaarTreeNode* haar_node_ptr;
192     CvSize orig_window_size;
193     int has_tilted_features = 0;
194     int max_count = 0;
195 
196     if( !CV_IS_HAAR_CLASSIFIER(cascade) )
197         CV_ERROR( !cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier pointer" );
198 
199     if( cascade->hid_cascade )
200         CV_ERROR( CV_StsError, "hid_cascade has been already created" );
201 
202     if( !cascade->stage_classifier )
203         CV_ERROR( CV_StsNullPtr, "" );
204 
205     if( cascade->count <= 0 )
206         CV_ERROR( CV_StsOutOfRange, "Negative number of cascade stages" );
207 
208     orig_window_size = cascade->orig_window_size;
209 
210     /* check input structure correctness and calculate total memory size needed for
211        internal representation of the classifier cascade */
212     for( i = 0; i < cascade->count; i++ )
213     {
214         CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
215 
216         if( !stage_classifier->classifier ||
217             stage_classifier->count <= 0 )
218         {
219             sprintf( errorstr, "header of the stage classifier #%d is invalid "
220                      "(has null pointers or non-positive classfier count)", i );
221             CV_ERROR( CV_StsError, errorstr );
222         }
223 
224         max_count = MAX( max_count, stage_classifier->count );
225         total_classifiers += stage_classifier->count;
226 
227         for( j = 0; j < stage_classifier->count; j++ )
228         {
229             CvHaarClassifier* classifier = stage_classifier->classifier + j;
230 
231             total_nodes += classifier->count;
232             for( l = 0; l < classifier->count; l++ )
233             {
234                 for( k = 0; k < CV_HAAR_FEATURE_MAX; k++ )
235                 {
236                     if( classifier->haar_feature[l].rect[k].r.width )
237                     {
238                         CvRect r = classifier->haar_feature[l].rect[k].r;
239                         int tilted = classifier->haar_feature[l].tilted;
240                         has_tilted_features |= tilted != 0;
241                         if( r.width < 0 || r.height < 0 || r.y < 0 ||
242                             r.x + r.width > orig_window_size.width
243                             ||
244                             (!tilted &&
245                             (r.x < 0 || r.y + r.height > orig_window_size.height))
246                             ||
247                             (tilted && (r.x - r.height < 0 ||
248                             r.y + r.width + r.height > orig_window_size.height)))
249                         {
250                             sprintf( errorstr, "rectangle #%d of the classifier #%d of "
251                                      "the stage classifier #%d is not inside "
252                                      "the reference (original) cascade window", k, j, i );
253                             CV_ERROR( CV_StsNullPtr, errorstr );
254                         }
255                     }
256                 }
257             }
258         }
259     }
260 
261     // this is an upper boundary for the whole hidden cascade size
262     datasize = sizeof(CvHidHaarClassifierCascade) +
263                sizeof(CvHidHaarStageClassifier)*cascade->count +
264                sizeof(CvHidHaarClassifier) * total_classifiers +
265                sizeof(CvHidHaarTreeNode) * total_nodes +
266                sizeof(void*)*(total_nodes + total_classifiers);
267 
268     CV_CALL( out = (CvHidHaarClassifierCascade*)cvAlloc( datasize ));
269     memset( out, 0, sizeof(*out) );
270 
271     /* init header */
272     out->count = cascade->count;
273     out->stage_classifier = (CvHidHaarStageClassifier*)(out + 1);
274     haar_classifier_ptr = (CvHidHaarClassifier*)(out->stage_classifier + cascade->count);
275     haar_node_ptr = (CvHidHaarTreeNode*)(haar_classifier_ptr + total_classifiers);
276 
277     out->is_stump_based = 1;
278     out->has_tilted_features = has_tilted_features;
279     out->is_tree = 0;
280 
281     /* initialize internal representation */
282     for( i = 0; i < cascade->count; i++ )
283     {
284         CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
285         CvHidHaarStageClassifier* hid_stage_classifier = out->stage_classifier + i;
286 
287         hid_stage_classifier->count = stage_classifier->count;
288         hid_stage_classifier->threshold = stage_classifier->threshold - icv_stage_threshold_bias;
289         hid_stage_classifier->classifier = haar_classifier_ptr;
290         hid_stage_classifier->two_rects = 1;
291         haar_classifier_ptr += stage_classifier->count;
292 
293         hid_stage_classifier->parent = (stage_classifier->parent == -1)
294             ? NULL : out->stage_classifier + stage_classifier->parent;
295         hid_stage_classifier->next = (stage_classifier->next == -1)
296             ? NULL : out->stage_classifier + stage_classifier->next;
297         hid_stage_classifier->child = (stage_classifier->child == -1)
298             ? NULL : out->stage_classifier + stage_classifier->child;
299 
300         out->is_tree |= hid_stage_classifier->next != NULL;
301 
302         for( j = 0; j < stage_classifier->count; j++ )
303         {
304             CvHaarClassifier* classifier = stage_classifier->classifier + j;
305             CvHidHaarClassifier* hid_classifier = hid_stage_classifier->classifier + j;
306             int node_count = classifier->count;
307             float* alpha_ptr = (float*)(haar_node_ptr + node_count);
308 
309             hid_classifier->count = node_count;
310             hid_classifier->node = haar_node_ptr;
311             hid_classifier->alpha = alpha_ptr;
312 
313             for( l = 0; l < node_count; l++ )
314             {
315                 CvHidHaarTreeNode* node = hid_classifier->node + l;
316                 CvHaarFeature* feature = classifier->haar_feature + l;
317                 memset( node, -1, sizeof(*node) );
318                 node->threshold = classifier->threshold[l];
319                 node->left = classifier->left[l];
320                 node->right = classifier->right[l];
321 
322                 if( fabs(feature->rect[2].weight) < DBL_EPSILON ||
323                     feature->rect[2].r.width == 0 ||
324                     feature->rect[2].r.height == 0 )
325                     memset( &(node->feature.rect[2]), 0, sizeof(node->feature.rect[2]) );
326                 else
327                     hid_stage_classifier->two_rects = 0;
328             }
329 
330             memcpy( alpha_ptr, classifier->alpha, (node_count+1)*sizeof(alpha_ptr[0]));
331             haar_node_ptr =
332                 (CvHidHaarTreeNode*)cvAlignPtr(alpha_ptr+node_count+1, sizeof(void*));
333 
334             out->is_stump_based &= node_count == 1;
335         }
336     }
337 
338     {
339     int can_use_ipp = icvHaarClassifierInitAlloc_32f_p != 0 &&
340         icvHaarClassifierFree_32f_p != 0 &&
341                       icvApplyHaarClassifier_32f_C1R_p != 0 &&
342                       icvRectStdDev_32f_C1R_p != 0 &&
343                       !out->has_tilted_features && !out->is_tree && out->is_stump_based;
344 
345     if( can_use_ipp )
346     {
347         int ipp_datasize = cascade->count*sizeof(out->ipp_stages[0]);
348         float ipp_weight_scale=(float)(1./((orig_window_size.width-icv_object_win_border*2)*
349             (orig_window_size.height-icv_object_win_border*2)));
350 
351         CV_CALL( out->ipp_stages = (void**)cvAlloc( ipp_datasize ));
352         memset( out->ipp_stages, 0, ipp_datasize );
353 
354         CV_CALL( ipp_features = (CvRect*)cvAlloc( max_count*3*sizeof(ipp_features[0]) ));
355         CV_CALL( ipp_weights = (float*)cvAlloc( max_count*3*sizeof(ipp_weights[0]) ));
356         CV_CALL( ipp_thresholds = (float*)cvAlloc( max_count*sizeof(ipp_thresholds[0]) ));
357         CV_CALL( ipp_val1 = (float*)cvAlloc( max_count*sizeof(ipp_val1[0]) ));
358         CV_CALL( ipp_val2 = (float*)cvAlloc( max_count*sizeof(ipp_val2[0]) ));
359         CV_CALL( ipp_counts = (int*)cvAlloc( max_count*sizeof(ipp_counts[0]) ));
360 
361         for( i = 0; i < cascade->count; i++ )
362         {
363             CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
364             for( j = 0, k = 0; j < stage_classifier->count; j++ )
365             {
366                 CvHaarClassifier* classifier = stage_classifier->classifier + j;
367                 int rect_count = 2 + (classifier->haar_feature->rect[2].r.width != 0);
368 
369                 ipp_thresholds[j] = classifier->threshold[0];
370                 ipp_val1[j] = classifier->alpha[0];
371                 ipp_val2[j] = classifier->alpha[1];
372                 ipp_counts[j] = rect_count;
373 
374                 for( l = 0; l < rect_count; l++, k++ )
375                 {
376                     ipp_features[k] = classifier->haar_feature->rect[l].r;
377                     //ipp_features[k].y = orig_window_size.height - ipp_features[k].y - ipp_features[k].height;
378                     ipp_weights[k] = classifier->haar_feature->rect[l].weight*ipp_weight_scale;
379                 }
380             }
381 
382             if( icvHaarClassifierInitAlloc_32f_p( &out->ipp_stages[i],
383                 ipp_features, ipp_weights, ipp_thresholds,
384                 ipp_val1, ipp_val2, ipp_counts, stage_classifier->count ) < 0 )
385                 break;
386         }
387 
388         if( i < cascade->count )
389         {
390             for( j = 0; j < i; j++ )
391                 if( icvHaarClassifierFree_32f_p && out->ipp_stages[i] )
392                     icvHaarClassifierFree_32f_p( out->ipp_stages[i] );
393             cvFree( &out->ipp_stages );
394         }
395     }
396     }
397 
398     cascade->hid_cascade = out;
399     assert( (char*)haar_node_ptr - (char*)out <= datasize );
400 
401     __END__;
402 
403     if( cvGetErrStatus() < 0 )
404         icvReleaseHidHaarClassifierCascade( &out );
405 
406     cvFree( &ipp_features );
407     cvFree( &ipp_weights );
408     cvFree( &ipp_thresholds );
409     cvFree( &ipp_val1 );
410     cvFree( &ipp_val2 );
411     cvFree( &ipp_counts );
412 
413     return out;
414 }
415 
416 
417 #define sum_elem_ptr(sum,row,col)  \
418     ((sumtype*)CV_MAT_ELEM_PTR_FAST((sum),(row),(col),sizeof(sumtype)))
419 
420 #define sqsum_elem_ptr(sqsum,row,col)  \
421     ((sqsumtype*)CV_MAT_ELEM_PTR_FAST((sqsum),(row),(col),sizeof(sqsumtype)))
422 
423 #define calc_sum(rect,offset) \
424     ((rect).p0[offset] - (rect).p1[offset] - (rect).p2[offset] + (rect).p3[offset])
425 
426 
427 CV_IMPL void
cvSetImagesForHaarClassifierCascade(CvHaarClassifierCascade * _cascade,const CvArr * _sum,const CvArr * _sqsum,const CvArr * _tilted_sum,double scale)428 cvSetImagesForHaarClassifierCascade( CvHaarClassifierCascade* _cascade,
429                                      const CvArr* _sum,
430                                      const CvArr* _sqsum,
431                                      const CvArr* _tilted_sum,
432                                      double scale )
433 {
434     CV_FUNCNAME("cvSetImagesForHaarClassifierCascade");
435 
436     __BEGIN__;
437 
438     CvMat sum_stub, *sum = (CvMat*)_sum;
439     CvMat sqsum_stub, *sqsum = (CvMat*)_sqsum;
440     CvMat tilted_stub, *tilted = (CvMat*)_tilted_sum;
441     CvHidHaarClassifierCascade* cascade;
442     int coi0 = 0, coi1 = 0;
443     int i;
444     CvRect equ_rect;
445     double weight_scale;
446 
447     if( !CV_IS_HAAR_CLASSIFIER(_cascade) )
448         CV_ERROR( !_cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier pointer" );
449 
450     if( scale <= 0 )
451         CV_ERROR( CV_StsOutOfRange, "Scale must be positive" );
452 
453     CV_CALL( sum = cvGetMat( sum, &sum_stub, &coi0 ));
454     CV_CALL( sqsum = cvGetMat( sqsum, &sqsum_stub, &coi1 ));
455 
456     if( coi0 || coi1 )
457         CV_ERROR( CV_BadCOI, "COI is not supported" );
458 
459     if( !CV_ARE_SIZES_EQ( sum, sqsum ))
460         CV_ERROR( CV_StsUnmatchedSizes, "All integral images must have the same size" );
461 
462     if( CV_MAT_TYPE(sqsum->type) != CV_64FC1 ||
463         CV_MAT_TYPE(sum->type) != CV_32SC1 )
464         CV_ERROR( CV_StsUnsupportedFormat,
465         "Only (32s, 64f, 32s) combination of (sum,sqsum,tilted_sum) formats is allowed" );
466 
467     if( !_cascade->hid_cascade )
468         CV_CALL( icvCreateHidHaarClassifierCascade(_cascade) );
469 
470     cascade = _cascade->hid_cascade;
471 
472     if( cascade->has_tilted_features )
473     {
474         CV_CALL( tilted = cvGetMat( tilted, &tilted_stub, &coi1 ));
475 
476         if( CV_MAT_TYPE(tilted->type) != CV_32SC1 )
477             CV_ERROR( CV_StsUnsupportedFormat,
478             "Only (32s, 64f, 32s) combination of (sum,sqsum,tilted_sum) formats is allowed" );
479 
480         if( sum->step != tilted->step )
481             CV_ERROR( CV_StsUnmatchedSizes,
482             "Sum and tilted_sum must have the same stride (step, widthStep)" );
483 
484         if( !CV_ARE_SIZES_EQ( sum, tilted ))
485             CV_ERROR( CV_StsUnmatchedSizes, "All integral images must have the same size" );
486         cascade->tilted = *tilted;
487     }
488 
489     _cascade->scale = scale;
490     _cascade->real_window_size.width = cvRound( _cascade->orig_window_size.width * scale );
491     _cascade->real_window_size.height = cvRound( _cascade->orig_window_size.height * scale );
492 
493     cascade->sum = *sum;
494     cascade->sqsum = *sqsum;
495 
496     equ_rect.x = equ_rect.y = cvRound(scale);
497     equ_rect.width = cvRound((_cascade->orig_window_size.width-2)*scale);
498     equ_rect.height = cvRound((_cascade->orig_window_size.height-2)*scale);
499     weight_scale = 1./(equ_rect.width*equ_rect.height);
500     cascade->inv_window_area = weight_scale;
501 
502     cascade->p0 = sum_elem_ptr(*sum, equ_rect.y, equ_rect.x);
503     cascade->p1 = sum_elem_ptr(*sum, equ_rect.y, equ_rect.x + equ_rect.width );
504     cascade->p2 = sum_elem_ptr(*sum, equ_rect.y + equ_rect.height, equ_rect.x );
505     cascade->p3 = sum_elem_ptr(*sum, equ_rect.y + equ_rect.height,
506                                      equ_rect.x + equ_rect.width );
507 
508     cascade->pq0 = sqsum_elem_ptr(*sqsum, equ_rect.y, equ_rect.x);
509     cascade->pq1 = sqsum_elem_ptr(*sqsum, equ_rect.y, equ_rect.x + equ_rect.width );
510     cascade->pq2 = sqsum_elem_ptr(*sqsum, equ_rect.y + equ_rect.height, equ_rect.x );
511     cascade->pq3 = sqsum_elem_ptr(*sqsum, equ_rect.y + equ_rect.height,
512                                           equ_rect.x + equ_rect.width );
513 
514     /* init pointers in haar features according to real window size and
515        given image pointers */
516     {
517 #ifdef _OPENMP
518     int max_threads = cvGetNumThreads();
519     #pragma omp parallel for num_threads(max_threads) schedule(dynamic)
520 #endif // _OPENMP
521     for( i = 0; i < _cascade->count; i++ )
522     {
523         int j, k, l;
524         for( j = 0; j < cascade->stage_classifier[i].count; j++ )
525         {
526             for( l = 0; l < cascade->stage_classifier[i].classifier[j].count; l++ )
527             {
528                 CvHaarFeature* feature =
529                     &_cascade->stage_classifier[i].classifier[j].haar_feature[l];
530                 /* CvHidHaarClassifier* classifier =
531                     cascade->stage_classifier[i].classifier + j; */
532                 CvHidHaarFeature* hidfeature =
533                     &cascade->stage_classifier[i].classifier[j].node[l].feature;
534                 double sum0 = 0, area0 = 0;
535                 CvRect r[3];
536 #if CV_ADJUST_FEATURES
537                 int base_w = -1, base_h = -1;
538                 int new_base_w = 0, new_base_h = 0;
539                 int kx, ky;
540                 int flagx = 0, flagy = 0;
541                 int x0 = 0, y0 = 0;
542 #endif
543                 int nr;
544 
545                 /* align blocks */
546                 for( k = 0; k < CV_HAAR_FEATURE_MAX; k++ )
547                 {
548                     if( !hidfeature->rect[k].p0 )
549                         break;
550 #if CV_ADJUST_FEATURES
551                     r[k] = feature->rect[k].r;
552                     base_w = (int)CV_IMIN( (unsigned)base_w, (unsigned)(r[k].width-1) );
553                     base_w = (int)CV_IMIN( (unsigned)base_w, (unsigned)(r[k].x - r[0].x-1) );
554                     base_h = (int)CV_IMIN( (unsigned)base_h, (unsigned)(r[k].height-1) );
555                     base_h = (int)CV_IMIN( (unsigned)base_h, (unsigned)(r[k].y - r[0].y-1) );
556 #endif
557                 }
558 
559                 nr = k;
560 
561 #if CV_ADJUST_FEATURES
562                 base_w += 1;
563                 base_h += 1;
564                 kx = r[0].width / base_w;
565                 ky = r[0].height / base_h;
566 
567                 if( kx <= 0 )
568                 {
569                     flagx = 1;
570                     new_base_w = cvRound( r[0].width * scale ) / kx;
571                     x0 = cvRound( r[0].x * scale );
572                 }
573 
574                 if( ky <= 0 )
575                 {
576                     flagy = 1;
577                     new_base_h = cvRound( r[0].height * scale ) / ky;
578                     y0 = cvRound( r[0].y * scale );
579                 }
580 #endif
581 
582                 for( k = 0; k < nr; k++ )
583                 {
584                     CvRect tr;
585                     double correction_ratio;
586 
587 #if CV_ADJUST_FEATURES
588                     if( flagx )
589                     {
590                         tr.x = (r[k].x - r[0].x) * new_base_w / base_w + x0;
591                         tr.width = r[k].width * new_base_w / base_w;
592                     }
593                     else
594 #endif
595                     {
596                         tr.x = cvRound( r[k].x * scale );
597                         tr.width = cvRound( r[k].width * scale );
598                     }
599 
600 #if CV_ADJUST_FEATURES
601                     if( flagy )
602                     {
603                         tr.y = (r[k].y - r[0].y) * new_base_h / base_h + y0;
604                         tr.height = r[k].height * new_base_h / base_h;
605                     }
606                     else
607 #endif
608                     {
609                         tr.y = cvRound( r[k].y * scale );
610                         tr.height = cvRound( r[k].height * scale );
611                     }
612 
613 #if CV_ADJUST_WEIGHTS
614                     {
615                     // RAINER START
616                     const float orig_feature_size =  (float)(feature->rect[k].r.width)*feature->rect[k].r.height;
617                     const float orig_norm_size = (float)(_cascade->orig_window_size.width)*(_cascade->orig_window_size.height);
618                     const float feature_size = float(tr.width*tr.height);
619                     //const float normSize    = float(equ_rect.width*equ_rect.height);
620                     float target_ratio = orig_feature_size / orig_norm_size;
621                     //float isRatio = featureSize / normSize;
622                     //correctionRatio = targetRatio / isRatio / normSize;
623                     correction_ratio = target_ratio / feature_size;
624                     // RAINER END
625                     }
626 #else
627                     correction_ratio = weight_scale * (!feature->tilted ? 1 : 0.5);
628 #endif
629 
630                     if( !feature->tilted )
631                     {
632                         hidfeature->rect[k].p0 = sum_elem_ptr(*sum, tr.y, tr.x);
633                         hidfeature->rect[k].p1 = sum_elem_ptr(*sum, tr.y, tr.x + tr.width);
634                         hidfeature->rect[k].p2 = sum_elem_ptr(*sum, tr.y + tr.height, tr.x);
635                         hidfeature->rect[k].p3 = sum_elem_ptr(*sum, tr.y + tr.height, tr.x + tr.width);
636                     }
637                     else
638                     {
639                         hidfeature->rect[k].p2 = sum_elem_ptr(*tilted, tr.y + tr.width, tr.x + tr.width);
640                         hidfeature->rect[k].p3 = sum_elem_ptr(*tilted, tr.y + tr.width + tr.height,
641                                                               tr.x + tr.width - tr.height);
642                         hidfeature->rect[k].p0 = sum_elem_ptr(*tilted, tr.y, tr.x);
643                         hidfeature->rect[k].p1 = sum_elem_ptr(*tilted, tr.y + tr.height, tr.x - tr.height);
644                     }
645 
646                     hidfeature->rect[k].weight = (float)(feature->rect[k].weight * correction_ratio);
647 
648                     if( k == 0 )
649                         area0 = tr.width * tr.height;
650                     else
651                         sum0 += hidfeature->rect[k].weight * tr.width * tr.height;
652                 }
653 
654                 hidfeature->rect[0].weight = (float)(-sum0/area0);
655             } /* l */
656         } /* j */
657     }
658     }
659 
660     __END__;
661 }
662 
663 
664 CV_INLINE
icvEvalHidHaarClassifier(CvHidHaarClassifier * classifier,double variance_norm_factor,size_t p_offset)665 double icvEvalHidHaarClassifier( CvHidHaarClassifier* classifier,
666                                  double variance_norm_factor,
667                                  size_t p_offset )
668 {
669     int idx = 0;
670     do
671     {
672         CvHidHaarTreeNode* node = classifier->node + idx;
673         double t = node->threshold * variance_norm_factor;
674 
675         double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
676         sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
677 
678         if( node->feature.rect[2].p0 )
679             sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
680 
681         idx = sum < t ? node->left : node->right;
682     }
683     while( idx > 0 );
684     return classifier->alpha[-idx];
685 }
686 
687 
688 CV_IMPL int
cvRunHaarClassifierCascade(CvHaarClassifierCascade * _cascade,CvPoint pt,int start_stage)689 cvRunHaarClassifierCascade( CvHaarClassifierCascade* _cascade,
690                             CvPoint pt, int start_stage )
691 {
692     int result = -1;
693     CV_FUNCNAME("cvRunHaarClassifierCascade");
694 
695     __BEGIN__;
696 
697     int p_offset, pq_offset;
698     int i, j;
699     double mean, variance_norm_factor;
700     CvHidHaarClassifierCascade* cascade;
701 
702     if( !CV_IS_HAAR_CLASSIFIER(_cascade) )
703         CV_ERROR( !_cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid cascade pointer" );
704 
705     cascade = _cascade->hid_cascade;
706     if( !cascade )
707         CV_ERROR( CV_StsNullPtr, "Hidden cascade has not been created.\n"
708             "Use cvSetImagesForHaarClassifierCascade" );
709 
710     if( pt.x < 0 || pt.y < 0 ||
711         pt.x + _cascade->real_window_size.width >= cascade->sum.width-2 ||
712         pt.y + _cascade->real_window_size.height >= cascade->sum.height-2 )
713         EXIT;
714 
715     p_offset = pt.y * (cascade->sum.step/sizeof(sumtype)) + pt.x;
716     pq_offset = pt.y * (cascade->sqsum.step/sizeof(sqsumtype)) + pt.x;
717     mean = calc_sum(*cascade,p_offset)*cascade->inv_window_area;
718     variance_norm_factor = cascade->pq0[pq_offset] - cascade->pq1[pq_offset] -
719                            cascade->pq2[pq_offset] + cascade->pq3[pq_offset];
720     variance_norm_factor = variance_norm_factor*cascade->inv_window_area - mean*mean;
721     if( variance_norm_factor >= 0. )
722         variance_norm_factor = sqrt(variance_norm_factor);
723     else
724         variance_norm_factor = 1.;
725 
726     if( cascade->is_tree )
727     {
728         CvHidHaarStageClassifier* ptr;
729         assert( start_stage == 0 );
730 
731         result = 1;
732         ptr = cascade->stage_classifier;
733 
734         while( ptr )
735         {
736             double stage_sum = 0;
737 
738             for( j = 0; j < ptr->count; j++ )
739             {
740                 stage_sum += icvEvalHidHaarClassifier( ptr->classifier + j,
741                     variance_norm_factor, p_offset );
742             }
743 
744             if( stage_sum >= ptr->threshold )
745             {
746                 ptr = ptr->child;
747             }
748             else
749             {
750                 while( ptr && ptr->next == NULL ) ptr = ptr->parent;
751                 if( ptr == NULL )
752                 {
753                     result = 0;
754                     EXIT;
755                 }
756                 ptr = ptr->next;
757             }
758         }
759     }
760     else if( cascade->is_stump_based )
761     {
762         for( i = start_stage; i < cascade->count; i++ )
763         {
764             double stage_sum = 0;
765 
766             if( cascade->stage_classifier[i].two_rects )
767             {
768                 for( j = 0; j < cascade->stage_classifier[i].count; j++ )
769                 {
770                     CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
771                     CvHidHaarTreeNode* node = classifier->node;
772                     double sum, t = node->threshold*variance_norm_factor, a, b;
773 
774                     sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
775                     sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
776 
777                     a = classifier->alpha[0];
778                     b = classifier->alpha[1];
779                     stage_sum += sum < t ? a : b;
780                 }
781             }
782             else
783             {
784                 for( j = 0; j < cascade->stage_classifier[i].count; j++ )
785                 {
786                     CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
787                     CvHidHaarTreeNode* node = classifier->node;
788                     double sum, t = node->threshold*variance_norm_factor, a, b;
789 
790                     sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
791                     sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
792 
793                     if( node->feature.rect[2].p0 )
794                         sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
795 
796                     a = classifier->alpha[0];
797                     b = classifier->alpha[1];
798                     stage_sum += sum < t ? a : b;
799                 }
800             }
801 
802             if( stage_sum < cascade->stage_classifier[i].threshold )
803             {
804                 result = -i;
805                 EXIT;
806             }
807         }
808     }
809     else
810     {
811         for( i = start_stage; i < cascade->count; i++ )
812         {
813             double stage_sum = 0;
814 
815             for( j = 0; j < cascade->stage_classifier[i].count; j++ )
816             {
817                 stage_sum += icvEvalHidHaarClassifier(
818                     cascade->stage_classifier[i].classifier + j,
819                     variance_norm_factor, p_offset );
820             }
821 
822             if( stage_sum < cascade->stage_classifier[i].threshold )
823             {
824                 result = -i;
825                 EXIT;
826             }
827         }
828     }
829 
830     result = 1;
831 
832     __END__;
833 
834     return result;
835 }
836 
837 
is_equal(const void * _r1,const void * _r2,void *)838 static int is_equal( const void* _r1, const void* _r2, void* )
839 {
840     const CvRect* r1 = (const CvRect*)_r1;
841     const CvRect* r2 = (const CvRect*)_r2;
842     int distance = cvRound(r1->width*0.2);
843 
844     return r2->x <= r1->x + distance &&
845            r2->x >= r1->x - distance &&
846            r2->y <= r1->y + distance &&
847            r2->y >= r1->y - distance &&
848            r2->width <= cvRound( r1->width * 1.2 ) &&
849            cvRound( r2->width * 1.2 ) >= r1->width;
850 }
851 
852 
853 #define VERY_ROUGH_SEARCH 0
854 
855 CV_IMPL CvSeq*
cvHaarDetectObjects(const CvArr * _img,CvHaarClassifierCascade * cascade,CvMemStorage * storage,double scale_factor,int min_neighbors,int flags,CvSize min_size)856 cvHaarDetectObjects( const CvArr* _img,
857                      CvHaarClassifierCascade* cascade,
858                      CvMemStorage* storage, double scale_factor,
859                      int min_neighbors, int flags, CvSize min_size )
860 {
861     int split_stage = 2;
862 
863     CvMat stub, *img = (CvMat*)_img;
864     CvMat *temp = 0, *sum = 0, *tilted = 0, *sqsum = 0, *norm_img = 0, *sumcanny = 0, *img_small = 0;
865     CvSeq* result_seq = 0;
866     CvMemStorage* temp_storage = 0;
867     CvAvgComp* comps = 0;
868     CvSeq* seq_thread[CV_MAX_THREADS] = {0};
869     int i, max_threads = 0;
870 
871     CV_FUNCNAME( "cvHaarDetectObjects" );
872 
873     __BEGIN__;
874 
875     CvSeq *seq = 0, *seq2 = 0, *idx_seq = 0, *big_seq = 0;
876     CvAvgComp result_comp = {{0,0,0,0},0};
877     double factor;
878     int npass = 2, coi;
879     bool do_canny_pruning = (flags & CV_HAAR_DO_CANNY_PRUNING) != 0;
880     bool find_biggest_object = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0;
881     bool rough_search = (flags & CV_HAAR_DO_ROUGH_SEARCH) != 0;
882 
883     if( !CV_IS_HAAR_CLASSIFIER(cascade) )
884         CV_ERROR( !cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier cascade" );
885 
886     if( !storage )
887         CV_ERROR( CV_StsNullPtr, "Null storage pointer" );
888 
889     CV_CALL( img = cvGetMat( img, &stub, &coi ));
890     if( coi )
891         CV_ERROR( CV_BadCOI, "COI is not supported" );
892 
893     if( CV_MAT_DEPTH(img->type) != CV_8U )
894         CV_ERROR( CV_StsUnsupportedFormat, "Only 8-bit images are supported" );
895 
896     if( find_biggest_object )
897         flags &= ~CV_HAAR_SCALE_IMAGE;
898 
899     CV_CALL( temp = cvCreateMat( img->rows, img->cols, CV_8UC1 ));
900     CV_CALL( sum = cvCreateMat( img->rows + 1, img->cols + 1, CV_32SC1 ));
901     CV_CALL( sqsum = cvCreateMat( img->rows + 1, img->cols + 1, CV_64FC1 ));
902     CV_CALL( temp_storage = cvCreateChildMemStorage( storage ));
903 
904     if( !cascade->hid_cascade )
905         CV_CALL( icvCreateHidHaarClassifierCascade(cascade) );
906 
907     if( cascade->hid_cascade->has_tilted_features )
908         tilted = cvCreateMat( img->rows + 1, img->cols + 1, CV_32SC1 );
909 
910     seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvRect), temp_storage );
911     seq2 = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), temp_storage );
912     result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), storage );
913 
914     max_threads = cvGetNumThreads();
915     if( max_threads > 1 )
916         for( i = 0; i < max_threads; i++ )
917         {
918             CvMemStorage* temp_storage_thread;
919             CV_CALL( temp_storage_thread = cvCreateMemStorage(0));
920             CV_CALL( seq_thread[i] = cvCreateSeq( 0, sizeof(CvSeq),
921                 sizeof(CvRect), temp_storage_thread ));
922         }
923     else
924         seq_thread[0] = seq;
925 
926     if( CV_MAT_CN(img->type) > 1 )
927     {
928         cvCvtColor( img, temp, CV_BGR2GRAY );
929         img = temp;
930     }
931 
932     if( flags & CV_HAAR_FIND_BIGGEST_OBJECT )
933         flags &= ~(CV_HAAR_SCALE_IMAGE|CV_HAAR_DO_CANNY_PRUNING);
934 
935     if( flags & CV_HAAR_SCALE_IMAGE )
936     {
937         CvSize win_size0 = cascade->orig_window_size;
938         int use_ipp = cascade->hid_cascade->ipp_stages != 0 &&
939                     icvApplyHaarClassifier_32f_C1R_p != 0;
940 
941         if( use_ipp )
942             CV_CALL( norm_img = cvCreateMat( img->rows, img->cols, CV_32FC1 ));
943         CV_CALL( img_small = cvCreateMat( img->rows + 1, img->cols + 1, CV_8UC1 ));
944 
945         for( factor = 1; ; factor *= scale_factor )
946         {
947             int strip_count, strip_size;
948             int ystep = factor > 2. ? 1 : 2;
949             CvSize win_size = { cvRound(win_size0.width*factor),
950                                 cvRound(win_size0.height*factor) };
951             CvSize sz = { cvRound( img->cols/factor ), cvRound( img->rows/factor ) };
952             CvSize sz1 = { sz.width - win_size0.width, sz.height - win_size0.height };
953             CvRect equ_rect = { icv_object_win_border, icv_object_win_border,
954                 win_size0.width - icv_object_win_border*2,
955                 win_size0.height - icv_object_win_border*2 };
956             CvMat img1, sum1, sqsum1, norm1, tilted1, mask1;
957             CvMat* _tilted = 0;
958 
959             if( sz1.width <= 0 || sz1.height <= 0 )
960                 break;
961             if( win_size.width < min_size.width || win_size.height < min_size.height )
962                 continue;
963 
964             img1 = cvMat( sz.height, sz.width, CV_8UC1, img_small->data.ptr );
965             sum1 = cvMat( sz.height+1, sz.width+1, CV_32SC1, sum->data.ptr );
966             sqsum1 = cvMat( sz.height+1, sz.width+1, CV_64FC1, sqsum->data.ptr );
967             if( tilted )
968             {
969                 tilted1 = cvMat( sz.height+1, sz.width+1, CV_32SC1, tilted->data.ptr );
970                 _tilted = &tilted1;
971             }
972             norm1 = cvMat( sz1.height, sz1.width, CV_32FC1, norm_img ? norm_img->data.ptr : 0 );
973             mask1 = cvMat( sz1.height, sz1.width, CV_8UC1, temp->data.ptr );
974 
975             cvResize( img, &img1, CV_INTER_LINEAR );
976             cvIntegral( &img1, &sum1, &sqsum1, _tilted );
977 
978             if( max_threads > 1 )
979             {
980                 strip_count = MAX(MIN(sz1.height/ystep, max_threads*3), 1);
981                 strip_size = (sz1.height + strip_count - 1)/strip_count;
982                 strip_size = (strip_size / ystep)*ystep;
983             }
984             else
985             {
986                 strip_count = 1;
987                 strip_size = sz1.height;
988             }
989 
990             if( !use_ipp )
991                 cvSetImagesForHaarClassifierCascade( cascade, &sum1, &sqsum1, 0, 1. );
992             else
993             {
994                 for( i = 0; i <= sz.height; i++ )
995                 {
996                     const int* isum = (int*)(sum1.data.ptr + sum1.step*i);
997                     float* fsum = (float*)isum;
998                     const int FLT_DELTA = -(1 << 24);
999                     int j;
1000                     for( j = 0; j <= sz.width; j++ )
1001                         fsum[j] = (float)(isum[j] + FLT_DELTA);
1002                 }
1003             }
1004 
1005         #ifdef _OPENMP
1006             #pragma omp parallel for num_threads(max_threads) schedule(dynamic)
1007         #endif
1008             for( i = 0; i < strip_count; i++ )
1009             {
1010                 int thread_id = cvGetThreadNum();
1011                 int positive = 0;
1012                 int y1 = i*strip_size, y2 = (i+1)*strip_size/* - ystep + 1*/;
1013                 CvSize ssz;
1014                 int x, y, j;
1015                 if( i == strip_count - 1 || y2 > sz1.height )
1016                     y2 = sz1.height;
1017                 ssz = cvSize(sz1.width, y2 - y1);
1018 
1019                 if( use_ipp )
1020                 {
1021                     icvRectStdDev_32f_C1R_p(
1022                         (float*)(sum1.data.ptr + y1*sum1.step), sum1.step,
1023                         (double*)(sqsum1.data.ptr + y1*sqsum1.step), sqsum1.step,
1024                         (float*)(norm1.data.ptr + y1*norm1.step), norm1.step, ssz, equ_rect );
1025 
1026                     positive = (ssz.width/ystep)*((ssz.height + ystep-1)/ystep);
1027                     memset( mask1.data.ptr + y1*mask1.step, ystep == 1, mask1.height*mask1.step);
1028 
1029                     if( ystep > 1 )
1030                     {
1031                         for( y = y1, positive = 0; y < y2; y += ystep )
1032                             for( x = 0; x < ssz.width; x += ystep )
1033                                 mask1.data.ptr[mask1.step*y + x] = (uchar)1;
1034                     }
1035 
1036                     for( j = 0; j < cascade->count; j++ )
1037                     {
1038                         if( icvApplyHaarClassifier_32f_C1R_p(
1039                             (float*)(sum1.data.ptr + y1*sum1.step), sum1.step,
1040                             (float*)(norm1.data.ptr + y1*norm1.step), norm1.step,
1041                             mask1.data.ptr + y1*mask1.step, mask1.step, ssz, &positive,
1042                             cascade->hid_cascade->stage_classifier[j].threshold,
1043                             cascade->hid_cascade->ipp_stages[j]) < 0 )
1044                         {
1045                             positive = 0;
1046                             break;
1047                         }
1048                         if( positive <= 0 )
1049                             break;
1050                     }
1051                 }
1052                 else
1053                 {
1054                     for( y = y1, positive = 0; y < y2; y += ystep )
1055                         for( x = 0; x < ssz.width; x += ystep )
1056                         {
1057                             mask1.data.ptr[mask1.step*y + x] =
1058                                 cvRunHaarClassifierCascade( cascade, cvPoint(x,y), 0 ) > 0;
1059                             positive += mask1.data.ptr[mask1.step*y + x];
1060                         }
1061                 }
1062 
1063                 if( positive > 0 )
1064                 {
1065                     for( y = y1; y < y2; y += ystep )
1066                         for( x = 0; x < ssz.width; x += ystep )
1067                             if( mask1.data.ptr[mask1.step*y + x] != 0 )
1068                             {
1069                                 CvRect obj_rect = { cvRound(x*factor), cvRound(y*factor),
1070                                                     win_size.width, win_size.height };
1071                                 cvSeqPush( seq_thread[thread_id], &obj_rect );
1072                             }
1073                 }
1074             }
1075 
1076             // gather the results
1077             if( max_threads > 1 )
1078                 for( i = 0; i < max_threads; i++ )
1079                 {
1080                     CvSeq* s = seq_thread[i];
1081                     int j, total = s->total;
1082                     CvSeqBlock* b = s->first;
1083                     for( j = 0; j < total; j += b->count, b = b->next )
1084                         cvSeqPushMulti( seq, b->data, b->count );
1085                 }
1086         }
1087     }
1088     else
1089     {
1090         int n_factors = 0;
1091         CvRect scan_roi_rect = {0,0,0,0};
1092         bool is_found = false, scan_roi = false;
1093 
1094         cvIntegral( img, sum, sqsum, tilted );
1095 
1096         if( do_canny_pruning )
1097         {
1098             sumcanny = cvCreateMat( img->rows + 1, img->cols + 1, CV_32SC1 );
1099             cvCanny( img, temp, 0, 50, 3 );
1100             cvIntegral( temp, sumcanny );
1101         }
1102 
1103         if( (unsigned)split_stage >= (unsigned)cascade->count ||
1104             cascade->hid_cascade->is_tree )
1105         {
1106             split_stage = cascade->count;
1107             npass = 1;
1108         }
1109 
1110         for( n_factors = 0, factor = 1;
1111              factor*cascade->orig_window_size.width < img->cols - 10 &&
1112              factor*cascade->orig_window_size.height < img->rows - 10;
1113              n_factors++, factor *= scale_factor )
1114             ;
1115 
1116         if( find_biggest_object )
1117         {
1118             scale_factor = 1./scale_factor;
1119             factor *= scale_factor;
1120             big_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvRect), temp_storage );
1121         }
1122         else
1123             factor = 1;
1124 
1125         for( ; n_factors-- > 0 && !is_found; factor *= scale_factor )
1126         {
1127             const double ystep = MAX( 2, factor );
1128             CvSize win_size = { cvRound( cascade->orig_window_size.width * factor ),
1129                                 cvRound( cascade->orig_window_size.height * factor )};
1130             CvRect equ_rect = { 0, 0, 0, 0 };
1131             int *p0 = 0, *p1 = 0, *p2 = 0, *p3 = 0;
1132             int *pq0 = 0, *pq1 = 0, *pq2 = 0, *pq3 = 0;
1133             int pass, stage_offset = 0;
1134             int start_x = 0, start_y = 0;
1135             int end_x = cvRound((img->cols - win_size.width) / ystep);
1136             int end_y = cvRound((img->rows - win_size.height) / ystep);
1137 
1138             if( win_size.width < min_size.width || win_size.height < min_size.height )
1139             {
1140                 if( find_biggest_object )
1141                     break;
1142                 continue;
1143             }
1144 
1145             cvSetImagesForHaarClassifierCascade( cascade, sum, sqsum, tilted, factor );
1146             cvZero( temp );
1147 
1148             if( do_canny_pruning )
1149             {
1150                 equ_rect.x = cvRound(win_size.width*0.15);
1151                 equ_rect.y = cvRound(win_size.height*0.15);
1152                 equ_rect.width = cvRound(win_size.width*0.7);
1153                 equ_rect.height = cvRound(win_size.height*0.7);
1154 
1155                 p0 = (int*)(sumcanny->data.ptr + equ_rect.y*sumcanny->step) + equ_rect.x;
1156                 p1 = (int*)(sumcanny->data.ptr + equ_rect.y*sumcanny->step)
1157                             + equ_rect.x + equ_rect.width;
1158                 p2 = (int*)(sumcanny->data.ptr + (equ_rect.y + equ_rect.height)*sumcanny->step) + equ_rect.x;
1159                 p3 = (int*)(sumcanny->data.ptr + (equ_rect.y + equ_rect.height)*sumcanny->step)
1160                             + equ_rect.x + equ_rect.width;
1161 
1162                 pq0 = (int*)(sum->data.ptr + equ_rect.y*sum->step) + equ_rect.x;
1163                 pq1 = (int*)(sum->data.ptr + equ_rect.y*sum->step)
1164                             + equ_rect.x + equ_rect.width;
1165                 pq2 = (int*)(sum->data.ptr + (equ_rect.y + equ_rect.height)*sum->step) + equ_rect.x;
1166                 pq3 = (int*)(sum->data.ptr + (equ_rect.y + equ_rect.height)*sum->step)
1167                             + equ_rect.x + equ_rect.width;
1168             }
1169 
1170             if( scan_roi )
1171             {
1172                 //adjust start_height and stop_height
1173                 start_y = cvRound(scan_roi_rect.y / ystep);
1174                 end_y = cvRound((scan_roi_rect.y + scan_roi_rect.height - win_size.height) / ystep);
1175 
1176                 start_x = cvRound(scan_roi_rect.x / ystep);
1177                 end_x = cvRound((scan_roi_rect.x + scan_roi_rect.width - win_size.width) / ystep);
1178             }
1179 
1180             cascade->hid_cascade->count = split_stage;
1181 
1182             for( pass = 0; pass < npass; pass++ )
1183             {
1184             #ifdef _OPENMP
1185                 #pragma omp parallel for num_threads(max_threads) schedule(dynamic)
1186             #endif
1187                 for( int _iy = start_y; _iy < end_y; _iy++ )
1188                 {
1189                     int thread_id = cvGetThreadNum();
1190                     int iy = cvRound(_iy*ystep);
1191                     int _ix, _xstep = 1;
1192                     uchar* mask_row = temp->data.ptr + temp->step * iy;
1193 
1194                     for( _ix = start_x; _ix < end_x; _ix += _xstep )
1195                     {
1196                         int ix = cvRound(_ix*ystep); // it really should be ystep
1197 
1198                         if( pass == 0 )
1199                         {
1200                             int result;
1201                             _xstep = 2;
1202 
1203                             if( do_canny_pruning )
1204                             {
1205                                 int offset;
1206                                 int s, sq;
1207 
1208                                 offset = iy*(sum->step/sizeof(p0[0])) + ix;
1209                                 s = p0[offset] - p1[offset] - p2[offset] + p3[offset];
1210                                 sq = pq0[offset] - pq1[offset] - pq2[offset] + pq3[offset];
1211                                 if( s < 100 || sq < 20 )
1212                                     continue;
1213                             }
1214 
1215                             result = cvRunHaarClassifierCascade( cascade, cvPoint(ix,iy), 0 );
1216                             if( result > 0 )
1217                             {
1218                                 if( pass < npass - 1 )
1219                                     mask_row[ix] = 1;
1220                                 else
1221                                 {
1222                                     CvRect rect = cvRect(ix,iy,win_size.width,win_size.height);
1223                                     cvSeqPush( seq_thread[thread_id], &rect );
1224                                 }
1225                             }
1226                             if( result < 0 )
1227                                 _xstep = 1;
1228                         }
1229                         else if( mask_row[ix] )
1230                         {
1231                             int result = cvRunHaarClassifierCascade( cascade, cvPoint(ix,iy),
1232                                                                      stage_offset );
1233                             if( result > 0 )
1234                             {
1235                                 if( pass == npass - 1 )
1236                                 {
1237                                     CvRect rect = cvRect(ix,iy,win_size.width,win_size.height);
1238                                     cvSeqPush( seq_thread[thread_id], &rect );
1239                                 }
1240                             }
1241                             else
1242                                 mask_row[ix] = 0;
1243                         }
1244                     }
1245                 }
1246                 stage_offset = cascade->hid_cascade->count;
1247                 cascade->hid_cascade->count = cascade->count;
1248             }
1249 
1250             // gather the results
1251             if( max_threads > 1 )
1252 	            for( i = 0; i < max_threads; i++ )
1253 	            {
1254 		            CvSeq* s = seq_thread[i];
1255                     int j, total = s->total;
1256                     CvSeqBlock* b = s->first;
1257                     for( j = 0; j < total; j += b->count, b = b->next )
1258                         cvSeqPushMulti( seq, b->data, b->count );
1259 	            }
1260 
1261             if( find_biggest_object )
1262             {
1263                 CvSeq* bseq = min_neighbors > 0 ? big_seq : seq;
1264 
1265                 if( min_neighbors > 0 && !scan_roi )
1266                 {
1267                     // group retrieved rectangles in order to filter out noise
1268                     int ncomp = cvSeqPartition( seq, 0, &idx_seq, is_equal, 0 );
1269                     CV_CALL( comps = (CvAvgComp*)cvAlloc( (ncomp+1)*sizeof(comps[0])));
1270                     memset( comps, 0, (ncomp+1)*sizeof(comps[0]));
1271 
1272                 #if VERY_ROUGH_SEARCH
1273                     if( rough_search )
1274                     {
1275                         for( i = 0; i < seq->total; i++ )
1276                         {
1277                             CvRect r1 = *(CvRect*)cvGetSeqElem( seq, i );
1278                             int idx = *(int*)cvGetSeqElem( idx_seq, i );
1279                             assert( (unsigned)idx < (unsigned)ncomp );
1280 
1281                             comps[idx].neighbors++;
1282                             comps[idx].rect.x += r1.x;
1283                             comps[idx].rect.y += r1.y;
1284                             comps[idx].rect.width += r1.width;
1285                             comps[idx].rect.height += r1.height;
1286                         }
1287 
1288                         // calculate average bounding box
1289                         for( i = 0; i < ncomp; i++ )
1290                         {
1291                             int n = comps[i].neighbors;
1292                             if( n >= min_neighbors )
1293                             {
1294                                 CvAvgComp comp;
1295                                 comp.rect.x = (comps[i].rect.x*2 + n)/(2*n);
1296                                 comp.rect.y = (comps[i].rect.y*2 + n)/(2*n);
1297                                 comp.rect.width = (comps[i].rect.width*2 + n)/(2*n);
1298                                 comp.rect.height = (comps[i].rect.height*2 + n)/(2*n);
1299                                 comp.neighbors = n;
1300                                 cvSeqPush( bseq, &comp );
1301                             }
1302                         }
1303                     }
1304                     else
1305                 #endif
1306                     {
1307                         for( i = 0 ; i <= ncomp; i++ )
1308                             comps[i].rect.x = comps[i].rect.y = INT_MAX;
1309 
1310                         // count number of neighbors
1311                         for( i = 0; i < seq->total; i++ )
1312                         {
1313                             CvRect r1 = *(CvRect*)cvGetSeqElem( seq, i );
1314                             int idx = *(int*)cvGetSeqElem( idx_seq, i );
1315                             assert( (unsigned)idx < (unsigned)ncomp );
1316 
1317                             comps[idx].neighbors++;
1318 
1319                             // rect.width and rect.height will store coordinate of right-bottom corner
1320                             comps[idx].rect.x = MIN(comps[idx].rect.x, r1.x);
1321                             comps[idx].rect.y = MIN(comps[idx].rect.y, r1.y);
1322                             comps[idx].rect.width = MAX(comps[idx].rect.width, r1.x+r1.width-1);
1323                             comps[idx].rect.height = MAX(comps[idx].rect.height, r1.y+r1.height-1);
1324                         }
1325 
1326                         // calculate enclosing box
1327                         for( i = 0; i < ncomp; i++ )
1328                         {
1329                             int n = comps[i].neighbors;
1330                             if( n >= min_neighbors )
1331                             {
1332                                 CvAvgComp comp;
1333                                 int t;
1334                                 double min_scale = rough_search ? 0.6 : 0.4;
1335                                 comp.rect.x = comps[i].rect.x;
1336                                 comp.rect.y = comps[i].rect.y;
1337                                 comp.rect.width = comps[i].rect.width - comps[i].rect.x + 1;
1338                                 comp.rect.height = comps[i].rect.height - comps[i].rect.y + 1;
1339 
1340                                 // update min_size
1341                                 t = cvRound( comp.rect.width*min_scale );
1342                                 min_size.width = MAX( min_size.width, t );
1343 
1344                                 t = cvRound( comp.rect.height*min_scale );
1345                                 min_size.height = MAX( min_size.height, t );
1346 
1347                                 //expand the box by 20% because we could miss some neighbours
1348                                 //see 'is_equal' function
1349                             #if 1
1350                                 int offset = cvRound(comp.rect.width * 0.2);
1351                                 int right = MIN( img->cols-1, comp.rect.x+comp.rect.width-1 + offset );
1352                                 int bottom = MIN( img->rows-1, comp.rect.y+comp.rect.height-1 + offset);
1353                                 comp.rect.x = MAX( comp.rect.x - offset, 0 );
1354                                 comp.rect.y = MAX( comp.rect.y - offset, 0 );
1355                                 comp.rect.width = right - comp.rect.x + 1;
1356                                 comp.rect.height = bottom - comp.rect.y + 1;
1357                             #endif
1358 
1359                                 comp.neighbors = n;
1360                                 cvSeqPush( bseq, &comp );
1361                             }
1362                         }
1363                     }
1364 
1365                     cvFree( &comps );
1366                 }
1367 
1368                 // extract the biggest rect
1369                 if( bseq->total > 0 )
1370                 {
1371                     int max_area = 0;
1372                     for( i = 0; i < bseq->total; i++ )
1373                     {
1374                         CvAvgComp* comp = (CvAvgComp*)cvGetSeqElem( bseq, i );
1375                         int area = comp->rect.width * comp->rect.height;
1376                         if( max_area < area )
1377                         {
1378                             max_area = area;
1379                             result_comp.rect = comp->rect;
1380                             result_comp.neighbors = bseq == seq ? 1 : comp->neighbors;
1381                         }
1382                     }
1383 
1384                     //Prepare information for further scanning inside the biggest rectangle
1385 
1386                 #if VERY_ROUGH_SEARCH
1387                     // change scan ranges to roi in case of required
1388                     if( !rough_search && !scan_roi )
1389                     {
1390                         scan_roi = true;
1391                         scan_roi_rect = result_comp.rect;
1392                         cvClearSeq(bseq);
1393                     }
1394                     else if( rough_search )
1395                         is_found = true;
1396                 #else
1397                     if( !scan_roi )
1398                     {
1399                         scan_roi = true;
1400                         scan_roi_rect = result_comp.rect;
1401                         cvClearSeq(bseq);
1402                     }
1403                 #endif
1404                 }
1405             }
1406         }
1407     }
1408 
1409     if( min_neighbors == 0 && !find_biggest_object )
1410     {
1411         for( i = 0; i < seq->total; i++ )
1412         {
1413             CvRect* rect = (CvRect*)cvGetSeqElem( seq, i );
1414             CvAvgComp comp;
1415             comp.rect = *rect;
1416             comp.neighbors = 1;
1417             cvSeqPush( result_seq, &comp );
1418         }
1419     }
1420 
1421     if( min_neighbors != 0
1422 #if VERY_ROUGH_SEARCH
1423         && (!find_biggest_object || !rough_search)
1424 #endif
1425         )
1426     {
1427         // group retrieved rectangles in order to filter out noise
1428         int ncomp = cvSeqPartition( seq, 0, &idx_seq, is_equal, 0 );
1429         CV_CALL( comps = (CvAvgComp*)cvAlloc( (ncomp+1)*sizeof(comps[0])));
1430         memset( comps, 0, (ncomp+1)*sizeof(comps[0]));
1431 
1432         // count number of neighbors
1433         for( i = 0; i < seq->total; i++ )
1434         {
1435             CvRect r1 = *(CvRect*)cvGetSeqElem( seq, i );
1436             int idx = *(int*)cvGetSeqElem( idx_seq, i );
1437             assert( (unsigned)idx < (unsigned)ncomp );
1438 
1439             comps[idx].neighbors++;
1440 
1441             comps[idx].rect.x += r1.x;
1442             comps[idx].rect.y += r1.y;
1443             comps[idx].rect.width += r1.width;
1444             comps[idx].rect.height += r1.height;
1445         }
1446 
1447         // calculate average bounding box
1448         for( i = 0; i < ncomp; i++ )
1449         {
1450             int n = comps[i].neighbors;
1451             if( n >= min_neighbors )
1452             {
1453                 CvAvgComp comp;
1454                 comp.rect.x = (comps[i].rect.x*2 + n)/(2*n);
1455                 comp.rect.y = (comps[i].rect.y*2 + n)/(2*n);
1456                 comp.rect.width = (comps[i].rect.width*2 + n)/(2*n);
1457                 comp.rect.height = (comps[i].rect.height*2 + n)/(2*n);
1458                 comp.neighbors = comps[i].neighbors;
1459 
1460                 cvSeqPush( seq2, &comp );
1461             }
1462         }
1463 
1464         if( !find_biggest_object )
1465         {
1466             // filter out small face rectangles inside large face rectangles
1467             for( i = 0; i < seq2->total; i++ )
1468             {
1469                 CvAvgComp r1 = *(CvAvgComp*)cvGetSeqElem( seq2, i );
1470                 int j, flag = 1;
1471 
1472                 for( j = 0; j < seq2->total; j++ )
1473                 {
1474                     CvAvgComp r2 = *(CvAvgComp*)cvGetSeqElem( seq2, j );
1475                     int distance = cvRound( r2.rect.width * 0.2 );
1476 
1477                     if( i != j &&
1478                         r1.rect.x >= r2.rect.x - distance &&
1479                         r1.rect.y >= r2.rect.y - distance &&
1480                         r1.rect.x + r1.rect.width <= r2.rect.x + r2.rect.width + distance &&
1481                         r1.rect.y + r1.rect.height <= r2.rect.y + r2.rect.height + distance &&
1482                         (r2.neighbors > MAX( 3, r1.neighbors ) || r1.neighbors < 3) )
1483                     {
1484                         flag = 0;
1485                         break;
1486                     }
1487                 }
1488 
1489                 if( flag )
1490                     cvSeqPush( result_seq, &r1 );
1491             }
1492         }
1493         else
1494         {
1495             int max_area = 0;
1496             for( i = 0; i < seq2->total; i++ )
1497             {
1498                 CvAvgComp* comp = (CvAvgComp*)cvGetSeqElem( seq2, i );
1499                 int area = comp->rect.width * comp->rect.height;
1500                 if( max_area < area )
1501                 {
1502                     max_area = area;
1503                     result_comp = *comp;
1504                 }
1505             }
1506         }
1507     }
1508 
1509     if( find_biggest_object && result_comp.rect.width > 0 )
1510         cvSeqPush( result_seq, &result_comp );
1511 
1512     __END__;
1513 
1514     if( max_threads > 1 )
1515 	    for( i = 0; i < max_threads; i++ )
1516 	    {
1517 		    if( seq_thread[i] )
1518                 cvReleaseMemStorage( &seq_thread[i]->storage );
1519 	    }
1520 
1521     cvReleaseMemStorage( &temp_storage );
1522     cvReleaseMat( &sum );
1523     cvReleaseMat( &sqsum );
1524     cvReleaseMat( &tilted );
1525     cvReleaseMat( &temp );
1526     cvReleaseMat( &sumcanny );
1527     cvReleaseMat( &norm_img );
1528     cvReleaseMat( &img_small );
1529     cvFree( &comps );
1530 
1531     return result_seq;
1532 }
1533 
1534 
1535 static CvHaarClassifierCascade*
icvLoadCascadeCART(const char ** input_cascade,int n,CvSize orig_window_size)1536 icvLoadCascadeCART( const char** input_cascade, int n, CvSize orig_window_size )
1537 {
1538     int i;
1539     CvHaarClassifierCascade* cascade = icvCreateHaarClassifierCascade(n);
1540     cascade->orig_window_size = orig_window_size;
1541 
1542     for( i = 0; i < n; i++ )
1543     {
1544         int j, count, l;
1545         float threshold = 0;
1546         const char* stage = input_cascade[i];
1547         int dl = 0;
1548 
1549         /* tree links */
1550         int parent = -1;
1551         int next = -1;
1552 
1553         sscanf( stage, "%d%n", &count, &dl );
1554         stage += dl;
1555 
1556         assert( count > 0 );
1557         cascade->stage_classifier[i].count = count;
1558         cascade->stage_classifier[i].classifier =
1559             (CvHaarClassifier*)cvAlloc( count*sizeof(cascade->stage_classifier[i].classifier[0]));
1560 
1561         for( j = 0; j < count; j++ )
1562         {
1563             CvHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
1564             int k, rects = 0;
1565             char str[100];
1566 
1567             sscanf( stage, "%d%n", &classifier->count, &dl );
1568             stage += dl;
1569 
1570             classifier->haar_feature = (CvHaarFeature*) cvAlloc(
1571                 classifier->count * ( sizeof( *classifier->haar_feature ) +
1572                                       sizeof( *classifier->threshold ) +
1573                                       sizeof( *classifier->left ) +
1574                                       sizeof( *classifier->right ) ) +
1575                 (classifier->count + 1) * sizeof( *classifier->alpha ) );
1576             classifier->threshold = (float*) (classifier->haar_feature+classifier->count);
1577             classifier->left = (int*) (classifier->threshold + classifier->count);
1578             classifier->right = (int*) (classifier->left + classifier->count);
1579             classifier->alpha = (float*) (classifier->right + classifier->count);
1580 
1581             for( l = 0; l < classifier->count; l++ )
1582             {
1583                 sscanf( stage, "%d%n", &rects, &dl );
1584                 stage += dl;
1585 
1586                 assert( rects >= 2 && rects <= CV_HAAR_FEATURE_MAX );
1587 
1588                 for( k = 0; k < rects; k++ )
1589                 {
1590                     CvRect r;
1591                     int band = 0;
1592                     sscanf( stage, "%d%d%d%d%d%f%n",
1593                             &r.x, &r.y, &r.width, &r.height, &band,
1594                             &(classifier->haar_feature[l].rect[k].weight), &dl );
1595                     stage += dl;
1596                     classifier->haar_feature[l].rect[k].r = r;
1597                 }
1598                 sscanf( stage, "%s%n", str, &dl );
1599                 stage += dl;
1600 
1601                 classifier->haar_feature[l].tilted = strncmp( str, "tilted", 6 ) == 0;
1602 
1603                 for( k = rects; k < CV_HAAR_FEATURE_MAX; k++ )
1604                 {
1605                     memset( classifier->haar_feature[l].rect + k, 0,
1606                             sizeof(classifier->haar_feature[l].rect[k]) );
1607                 }
1608 
1609                 sscanf( stage, "%f%d%d%n", &(classifier->threshold[l]),
1610                                        &(classifier->left[l]),
1611                                        &(classifier->right[l]), &dl );
1612                 stage += dl;
1613             }
1614             for( l = 0; l <= classifier->count; l++ )
1615             {
1616                 sscanf( stage, "%f%n", &(classifier->alpha[l]), &dl );
1617                 stage += dl;
1618             }
1619         }
1620 
1621         sscanf( stage, "%f%n", &threshold, &dl );
1622         stage += dl;
1623 
1624         cascade->stage_classifier[i].threshold = threshold;
1625 
1626         /* load tree links */
1627         if( sscanf( stage, "%d%d%n", &parent, &next, &dl ) != 2 )
1628         {
1629             parent = i - 1;
1630             next = -1;
1631         }
1632         stage += dl;
1633 
1634         cascade->stage_classifier[i].parent = parent;
1635         cascade->stage_classifier[i].next = next;
1636         cascade->stage_classifier[i].child = -1;
1637 
1638         if( parent != -1 && cascade->stage_classifier[parent].child == -1 )
1639         {
1640             cascade->stage_classifier[parent].child = i;
1641         }
1642     }
1643 
1644     return cascade;
1645 }
1646 
1647 #ifndef _MAX_PATH
1648 #define _MAX_PATH 1024
1649 #endif
1650 
1651 CV_IMPL CvHaarClassifierCascade*
cvLoadHaarClassifierCascade(const char * directory,CvSize orig_window_size)1652 cvLoadHaarClassifierCascade( const char* directory, CvSize orig_window_size )
1653 {
1654     const char** input_cascade = 0;
1655     CvHaarClassifierCascade *cascade = 0;
1656 
1657     CV_FUNCNAME( "cvLoadHaarClassifierCascade" );
1658 
1659     __BEGIN__;
1660 
1661     int i, n;
1662     const char* slash;
1663     char name[_MAX_PATH];
1664     int size = 0;
1665     char* ptr = 0;
1666 
1667     if( !directory )
1668         CV_ERROR( CV_StsNullPtr, "Null path is passed" );
1669 
1670     n = (int)strlen(directory)-1;
1671     slash = directory[n] == '\\' || directory[n] == '/' ? "" : "/";
1672 
1673     /* try to read the classifier from directory */
1674     for( n = 0; ; n++ )
1675     {
1676         sprintf( name, "%s%s%d/AdaBoostCARTHaarClassifier.txt", directory, slash, n );
1677         FILE* f = fopen( name, "rb" );
1678         if( !f )
1679             break;
1680         fseek( f, 0, SEEK_END );
1681         size += ftell( f ) + 1;
1682         fclose(f);
1683     }
1684 
1685     if( n == 0 && slash[0] )
1686     {
1687         CV_CALL( cascade = (CvHaarClassifierCascade*)cvLoad( directory ));
1688         EXIT;
1689     }
1690     else if( n == 0 )
1691         CV_ERROR( CV_StsBadArg, "Invalid path" );
1692 
1693     size += (n+1)*sizeof(char*);
1694     CV_CALL( input_cascade = (const char**)cvAlloc( size ));
1695     ptr = (char*)(input_cascade + n + 1);
1696 
1697     for( i = 0; i < n; i++ )
1698     {
1699         sprintf( name, "%s/%d/AdaBoostCARTHaarClassifier.txt", directory, i );
1700         FILE* f = fopen( name, "rb" );
1701         if( !f )
1702             CV_ERROR( CV_StsError, "" );
1703         fseek( f, 0, SEEK_END );
1704         size = ftell( f );
1705         fseek( f, 0, SEEK_SET );
1706         fread( ptr, 1, size, f );
1707         fclose(f);
1708         input_cascade[i] = ptr;
1709         ptr += size;
1710         *ptr++ = '\0';
1711     }
1712 
1713     input_cascade[n] = 0;
1714     cascade = icvLoadCascadeCART( input_cascade, n, orig_window_size );
1715 
1716     __END__;
1717 
1718     if( input_cascade )
1719         cvFree( &input_cascade );
1720 
1721     if( cvGetErrStatus() < 0 )
1722         cvReleaseHaarClassifierCascade( &cascade );
1723 
1724     return cascade;
1725 }
1726 
1727 
1728 CV_IMPL void
cvReleaseHaarClassifierCascade(CvHaarClassifierCascade ** _cascade)1729 cvReleaseHaarClassifierCascade( CvHaarClassifierCascade** _cascade )
1730 {
1731     if( _cascade && *_cascade )
1732     {
1733         int i, j;
1734         CvHaarClassifierCascade* cascade = *_cascade;
1735 
1736         for( i = 0; i < cascade->count; i++ )
1737         {
1738             for( j = 0; j < cascade->stage_classifier[i].count; j++ )
1739                 cvFree( &cascade->stage_classifier[i].classifier[j].haar_feature );
1740             cvFree( &cascade->stage_classifier[i].classifier );
1741         }
1742         icvReleaseHidHaarClassifierCascade( &cascade->hid_cascade );
1743         cvFree( _cascade );
1744     }
1745 }
1746 
1747 
1748 /****************************************************************************************\
1749 *                                  Persistence functions                                 *
1750 \****************************************************************************************/
1751 
1752 /* field names */
1753 
1754 #define ICV_HAAR_SIZE_NAME            "size"
1755 #define ICV_HAAR_STAGES_NAME          "stages"
1756 #define ICV_HAAR_TREES_NAME             "trees"
1757 #define ICV_HAAR_FEATURE_NAME             "feature"
1758 #define ICV_HAAR_RECTS_NAME                 "rects"
1759 #define ICV_HAAR_TILTED_NAME                "tilted"
1760 #define ICV_HAAR_THRESHOLD_NAME           "threshold"
1761 #define ICV_HAAR_LEFT_NODE_NAME           "left_node"
1762 #define ICV_HAAR_LEFT_VAL_NAME            "left_val"
1763 #define ICV_HAAR_RIGHT_NODE_NAME          "right_node"
1764 #define ICV_HAAR_RIGHT_VAL_NAME           "right_val"
1765 #define ICV_HAAR_STAGE_THRESHOLD_NAME   "stage_threshold"
1766 #define ICV_HAAR_PARENT_NAME            "parent"
1767 #define ICV_HAAR_NEXT_NAME              "next"
1768 
1769 static int
icvIsHaarClassifier(const void * struct_ptr)1770 icvIsHaarClassifier( const void* struct_ptr )
1771 {
1772     return CV_IS_HAAR_CLASSIFIER( struct_ptr );
1773 }
1774 
1775 static void*
icvReadHaarClassifier(CvFileStorage * fs,CvFileNode * node)1776 icvReadHaarClassifier( CvFileStorage* fs, CvFileNode* node )
1777 {
1778     CvHaarClassifierCascade* cascade = NULL;
1779 
1780     CV_FUNCNAME( "cvReadHaarClassifier" );
1781 
1782     __BEGIN__;
1783 
1784     char buf[256];
1785     CvFileNode* seq_fn = NULL; /* sequence */
1786     CvFileNode* fn = NULL;
1787     CvFileNode* stages_fn = NULL;
1788     CvSeqReader stages_reader;
1789     int n;
1790     int i, j, k, l;
1791     int parent, next;
1792 
1793     CV_CALL( stages_fn = cvGetFileNodeByName( fs, node, ICV_HAAR_STAGES_NAME ) );
1794     if( !stages_fn || !CV_NODE_IS_SEQ( stages_fn->tag) )
1795         CV_ERROR( CV_StsError, "Invalid stages node" );
1796 
1797     n = stages_fn->data.seq->total;
1798     CV_CALL( cascade = icvCreateHaarClassifierCascade(n) );
1799 
1800     /* read size */
1801     CV_CALL( seq_fn = cvGetFileNodeByName( fs, node, ICV_HAAR_SIZE_NAME ) );
1802     if( !seq_fn || !CV_NODE_IS_SEQ( seq_fn->tag ) || seq_fn->data.seq->total != 2 )
1803         CV_ERROR( CV_StsError, "size node is not a valid sequence." );
1804     CV_CALL( fn = (CvFileNode*) cvGetSeqElem( seq_fn->data.seq, 0 ) );
1805     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= 0 )
1806         CV_ERROR( CV_StsError, "Invalid size node: width must be positive integer" );
1807     cascade->orig_window_size.width = fn->data.i;
1808     CV_CALL( fn = (CvFileNode*) cvGetSeqElem( seq_fn->data.seq, 1 ) );
1809     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= 0 )
1810         CV_ERROR( CV_StsError, "Invalid size node: height must be positive integer" );
1811     cascade->orig_window_size.height = fn->data.i;
1812 
1813     CV_CALL( cvStartReadSeq( stages_fn->data.seq, &stages_reader ) );
1814     for( i = 0; i < n; ++i )
1815     {
1816         CvFileNode* stage_fn;
1817         CvFileNode* trees_fn;
1818         CvSeqReader trees_reader;
1819 
1820         stage_fn = (CvFileNode*) stages_reader.ptr;
1821         if( !CV_NODE_IS_MAP( stage_fn->tag ) )
1822         {
1823             sprintf( buf, "Invalid stage %d", i );
1824             CV_ERROR( CV_StsError, buf );
1825         }
1826 
1827         CV_CALL( trees_fn = cvGetFileNodeByName( fs, stage_fn, ICV_HAAR_TREES_NAME ) );
1828         if( !trees_fn || !CV_NODE_IS_SEQ( trees_fn->tag )
1829             || trees_fn->data.seq->total <= 0 )
1830         {
1831             sprintf( buf, "Trees node is not a valid sequence. (stage %d)", i );
1832             CV_ERROR( CV_StsError, buf );
1833         }
1834 
1835         CV_CALL( cascade->stage_classifier[i].classifier =
1836             (CvHaarClassifier*) cvAlloc( trees_fn->data.seq->total
1837                 * sizeof( cascade->stage_classifier[i].classifier[0] ) ) );
1838         for( j = 0; j < trees_fn->data.seq->total; ++j )
1839         {
1840             cascade->stage_classifier[i].classifier[j].haar_feature = NULL;
1841         }
1842         cascade->stage_classifier[i].count = trees_fn->data.seq->total;
1843 
1844         CV_CALL( cvStartReadSeq( trees_fn->data.seq, &trees_reader ) );
1845         for( j = 0; j < trees_fn->data.seq->total; ++j )
1846         {
1847             CvFileNode* tree_fn;
1848             CvSeqReader tree_reader;
1849             CvHaarClassifier* classifier;
1850             int last_idx;
1851 
1852             classifier = &cascade->stage_classifier[i].classifier[j];
1853             tree_fn = (CvFileNode*) trees_reader.ptr;
1854             if( !CV_NODE_IS_SEQ( tree_fn->tag ) || tree_fn->data.seq->total <= 0 )
1855             {
1856                 sprintf( buf, "Tree node is not a valid sequence."
1857                          " (stage %d, tree %d)", i, j );
1858                 CV_ERROR( CV_StsError, buf );
1859             }
1860 
1861             classifier->count = tree_fn->data.seq->total;
1862             CV_CALL( classifier->haar_feature = (CvHaarFeature*) cvAlloc(
1863                 classifier->count * ( sizeof( *classifier->haar_feature ) +
1864                                       sizeof( *classifier->threshold ) +
1865                                       sizeof( *classifier->left ) +
1866                                       sizeof( *classifier->right ) ) +
1867                 (classifier->count + 1) * sizeof( *classifier->alpha ) ) );
1868             classifier->threshold = (float*) (classifier->haar_feature+classifier->count);
1869             classifier->left = (int*) (classifier->threshold + classifier->count);
1870             classifier->right = (int*) (classifier->left + classifier->count);
1871             classifier->alpha = (float*) (classifier->right + classifier->count);
1872 
1873             CV_CALL( cvStartReadSeq( tree_fn->data.seq, &tree_reader ) );
1874             for( k = 0, last_idx = 0; k < tree_fn->data.seq->total; ++k )
1875             {
1876                 CvFileNode* node_fn;
1877                 CvFileNode* feature_fn;
1878                 CvFileNode* rects_fn;
1879                 CvSeqReader rects_reader;
1880 
1881                 node_fn = (CvFileNode*) tree_reader.ptr;
1882                 if( !CV_NODE_IS_MAP( node_fn->tag ) )
1883                 {
1884                     sprintf( buf, "Tree node %d is not a valid map. (stage %d, tree %d)",
1885                              k, i, j );
1886                     CV_ERROR( CV_StsError, buf );
1887                 }
1888                 CV_CALL( feature_fn = cvGetFileNodeByName( fs, node_fn,
1889                     ICV_HAAR_FEATURE_NAME ) );
1890                 if( !feature_fn || !CV_NODE_IS_MAP( feature_fn->tag ) )
1891                 {
1892                     sprintf( buf, "Feature node is not a valid map. "
1893                              "(stage %d, tree %d, node %d)", i, j, k );
1894                     CV_ERROR( CV_StsError, buf );
1895                 }
1896                 CV_CALL( rects_fn = cvGetFileNodeByName( fs, feature_fn,
1897                     ICV_HAAR_RECTS_NAME ) );
1898                 if( !rects_fn || !CV_NODE_IS_SEQ( rects_fn->tag )
1899                     || rects_fn->data.seq->total < 1
1900                     || rects_fn->data.seq->total > CV_HAAR_FEATURE_MAX )
1901                 {
1902                     sprintf( buf, "Rects node is not a valid sequence. "
1903                              "(stage %d, tree %d, node %d)", i, j, k );
1904                     CV_ERROR( CV_StsError, buf );
1905                 }
1906                 CV_CALL( cvStartReadSeq( rects_fn->data.seq, &rects_reader ) );
1907                 for( l = 0; l < rects_fn->data.seq->total; ++l )
1908                 {
1909                     CvFileNode* rect_fn;
1910                     CvRect r;
1911 
1912                     rect_fn = (CvFileNode*) rects_reader.ptr;
1913                     if( !CV_NODE_IS_SEQ( rect_fn->tag ) || rect_fn->data.seq->total != 5 )
1914                     {
1915                         sprintf( buf, "Rect %d is not a valid sequence. "
1916                                  "(stage %d, tree %d, node %d)", l, i, j, k );
1917                         CV_ERROR( CV_StsError, buf );
1918                     }
1919 
1920                     fn = CV_SEQ_ELEM( rect_fn->data.seq, CvFileNode, 0 );
1921                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i < 0 )
1922                     {
1923                         sprintf( buf, "x coordinate must be non-negative integer. "
1924                                  "(stage %d, tree %d, node %d, rect %d)", i, j, k, l );
1925                         CV_ERROR( CV_StsError, buf );
1926                     }
1927                     r.x = fn->data.i;
1928                     fn = CV_SEQ_ELEM( rect_fn->data.seq, CvFileNode, 1 );
1929                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i < 0 )
1930                     {
1931                         sprintf( buf, "y coordinate must be non-negative integer. "
1932                                  "(stage %d, tree %d, node %d, rect %d)", i, j, k, l );
1933                         CV_ERROR( CV_StsError, buf );
1934                     }
1935                     r.y = fn->data.i;
1936                     fn = CV_SEQ_ELEM( rect_fn->data.seq, CvFileNode, 2 );
1937                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= 0
1938                         || r.x + fn->data.i > cascade->orig_window_size.width )
1939                     {
1940                         sprintf( buf, "width must be positive integer and "
1941                                  "(x + width) must not exceed window width. "
1942                                  "(stage %d, tree %d, node %d, rect %d)", i, j, k, l );
1943                         CV_ERROR( CV_StsError, buf );
1944                     }
1945                     r.width = fn->data.i;
1946                     fn = CV_SEQ_ELEM( rect_fn->data.seq, CvFileNode, 3 );
1947                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= 0
1948                         || r.y + fn->data.i > cascade->orig_window_size.height )
1949                     {
1950                         sprintf( buf, "height must be positive integer and "
1951                                  "(y + height) must not exceed window height. "
1952                                  "(stage %d, tree %d, node %d, rect %d)", i, j, k, l );
1953                         CV_ERROR( CV_StsError, buf );
1954                     }
1955                     r.height = fn->data.i;
1956                     fn = CV_SEQ_ELEM( rect_fn->data.seq, CvFileNode, 4 );
1957                     if( !CV_NODE_IS_REAL( fn->tag ) )
1958                     {
1959                         sprintf( buf, "weight must be real number. "
1960                                  "(stage %d, tree %d, node %d, rect %d)", i, j, k, l );
1961                         CV_ERROR( CV_StsError, buf );
1962                     }
1963 
1964                     classifier->haar_feature[k].rect[l].weight = (float) fn->data.f;
1965                     classifier->haar_feature[k].rect[l].r = r;
1966 
1967                     CV_NEXT_SEQ_ELEM( sizeof( *rect_fn ), rects_reader );
1968                 } /* for each rect */
1969                 for( l = rects_fn->data.seq->total; l < CV_HAAR_FEATURE_MAX; ++l )
1970                 {
1971                     classifier->haar_feature[k].rect[l].weight = 0;
1972                     classifier->haar_feature[k].rect[l].r = cvRect( 0, 0, 0, 0 );
1973                 }
1974 
1975                 CV_CALL( fn = cvGetFileNodeByName( fs, feature_fn, ICV_HAAR_TILTED_NAME));
1976                 if( !fn || !CV_NODE_IS_INT( fn->tag ) )
1977                 {
1978                     sprintf( buf, "tilted must be 0 or 1. "
1979                              "(stage %d, tree %d, node %d)", i, j, k );
1980                     CV_ERROR( CV_StsError, buf );
1981                 }
1982                 classifier->haar_feature[k].tilted = ( fn->data.i != 0 );
1983                 CV_CALL( fn = cvGetFileNodeByName( fs, node_fn, ICV_HAAR_THRESHOLD_NAME));
1984                 if( !fn || !CV_NODE_IS_REAL( fn->tag ) )
1985                 {
1986                     sprintf( buf, "threshold must be real number. "
1987                              "(stage %d, tree %d, node %d)", i, j, k );
1988                     CV_ERROR( CV_StsError, buf );
1989                 }
1990                 classifier->threshold[k] = (float) fn->data.f;
1991                 CV_CALL( fn = cvGetFileNodeByName( fs, node_fn, ICV_HAAR_LEFT_NODE_NAME));
1992                 if( fn )
1993                 {
1994                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= k
1995                         || fn->data.i >= tree_fn->data.seq->total )
1996                     {
1997                         sprintf( buf, "left node must be valid node number. "
1998                                  "(stage %d, tree %d, node %d)", i, j, k );
1999                         CV_ERROR( CV_StsError, buf );
2000                     }
2001                     /* left node */
2002                     classifier->left[k] = fn->data.i;
2003                 }
2004                 else
2005                 {
2006                     CV_CALL( fn = cvGetFileNodeByName( fs, node_fn,
2007                         ICV_HAAR_LEFT_VAL_NAME ) );
2008                     if( !fn )
2009                     {
2010                         sprintf( buf, "left node or left value must be specified. "
2011                                  "(stage %d, tree %d, node %d)", i, j, k );
2012                         CV_ERROR( CV_StsError, buf );
2013                     }
2014                     if( !CV_NODE_IS_REAL( fn->tag ) )
2015                     {
2016                         sprintf( buf, "left value must be real number. "
2017                                  "(stage %d, tree %d, node %d)", i, j, k );
2018                         CV_ERROR( CV_StsError, buf );
2019                     }
2020                     /* left value */
2021                     if( last_idx >= classifier->count + 1 )
2022                     {
2023                         sprintf( buf, "Tree structure is broken: too many values. "
2024                                  "(stage %d, tree %d, node %d)", i, j, k );
2025                         CV_ERROR( CV_StsError, buf );
2026                     }
2027                     classifier->left[k] = -last_idx;
2028                     classifier->alpha[last_idx++] = (float) fn->data.f;
2029                 }
2030                 CV_CALL( fn = cvGetFileNodeByName( fs, node_fn,ICV_HAAR_RIGHT_NODE_NAME));
2031                 if( fn )
2032                 {
2033                     if( !CV_NODE_IS_INT( fn->tag ) || fn->data.i <= k
2034                         || fn->data.i >= tree_fn->data.seq->total )
2035                     {
2036                         sprintf( buf, "right node must be valid node number. "
2037                                  "(stage %d, tree %d, node %d)", i, j, k );
2038                         CV_ERROR( CV_StsError, buf );
2039                     }
2040                     /* right node */
2041                     classifier->right[k] = fn->data.i;
2042                 }
2043                 else
2044                 {
2045                     CV_CALL( fn = cvGetFileNodeByName( fs, node_fn,
2046                         ICV_HAAR_RIGHT_VAL_NAME ) );
2047                     if( !fn )
2048                     {
2049                         sprintf( buf, "right node or right value must be specified. "
2050                                  "(stage %d, tree %d, node %d)", i, j, k );
2051                         CV_ERROR( CV_StsError, buf );
2052                     }
2053                     if( !CV_NODE_IS_REAL( fn->tag ) )
2054                     {
2055                         sprintf( buf, "right value must be real number. "
2056                                  "(stage %d, tree %d, node %d)", i, j, k );
2057                         CV_ERROR( CV_StsError, buf );
2058                     }
2059                     /* right value */
2060                     if( last_idx >= classifier->count + 1 )
2061                     {
2062                         sprintf( buf, "Tree structure is broken: too many values. "
2063                                  "(stage %d, tree %d, node %d)", i, j, k );
2064                         CV_ERROR( CV_StsError, buf );
2065                     }
2066                     classifier->right[k] = -last_idx;
2067                     classifier->alpha[last_idx++] = (float) fn->data.f;
2068                 }
2069 
2070                 CV_NEXT_SEQ_ELEM( sizeof( *node_fn ), tree_reader );
2071             } /* for each node */
2072             if( last_idx != classifier->count + 1 )
2073             {
2074                 sprintf( buf, "Tree structure is broken: too few values. "
2075                          "(stage %d, tree %d)", i, j );
2076                 CV_ERROR( CV_StsError, buf );
2077             }
2078 
2079             CV_NEXT_SEQ_ELEM( sizeof( *tree_fn ), trees_reader );
2080         } /* for each tree */
2081 
2082         CV_CALL( fn = cvGetFileNodeByName( fs, stage_fn, ICV_HAAR_STAGE_THRESHOLD_NAME));
2083         if( !fn || !CV_NODE_IS_REAL( fn->tag ) )
2084         {
2085             sprintf( buf, "stage threshold must be real number. (stage %d)", i );
2086             CV_ERROR( CV_StsError, buf );
2087         }
2088         cascade->stage_classifier[i].threshold = (float) fn->data.f;
2089 
2090         parent = i - 1;
2091         next = -1;
2092 
2093         CV_CALL( fn = cvGetFileNodeByName( fs, stage_fn, ICV_HAAR_PARENT_NAME ) );
2094         if( !fn || !CV_NODE_IS_INT( fn->tag )
2095             || fn->data.i < -1 || fn->data.i >= cascade->count )
2096         {
2097             sprintf( buf, "parent must be integer number. (stage %d)", i );
2098             CV_ERROR( CV_StsError, buf );
2099         }
2100         parent = fn->data.i;
2101         CV_CALL( fn = cvGetFileNodeByName( fs, stage_fn, ICV_HAAR_NEXT_NAME ) );
2102         if( !fn || !CV_NODE_IS_INT( fn->tag )
2103             || fn->data.i < -1 || fn->data.i >= cascade->count )
2104         {
2105             sprintf( buf, "next must be integer number. (stage %d)", i );
2106             CV_ERROR( CV_StsError, buf );
2107         }
2108         next = fn->data.i;
2109 
2110         cascade->stage_classifier[i].parent = parent;
2111         cascade->stage_classifier[i].next = next;
2112         cascade->stage_classifier[i].child = -1;
2113 
2114         if( parent != -1 && cascade->stage_classifier[parent].child == -1 )
2115         {
2116             cascade->stage_classifier[parent].child = i;
2117         }
2118 
2119         CV_NEXT_SEQ_ELEM( sizeof( *stage_fn ), stages_reader );
2120     } /* for each stage */
2121 
2122     __END__;
2123 
2124     if( cvGetErrStatus() < 0 )
2125     {
2126         cvReleaseHaarClassifierCascade( &cascade );
2127         cascade = NULL;
2128     }
2129 
2130     return cascade;
2131 }
2132 
2133 static void
icvWriteHaarClassifier(CvFileStorage * fs,const char * name,const void * struct_ptr,CvAttrList attributes)2134 icvWriteHaarClassifier( CvFileStorage* fs, const char* name, const void* struct_ptr,
2135                         CvAttrList attributes )
2136 {
2137     CV_FUNCNAME( "cvWriteHaarClassifier" );
2138 
2139     __BEGIN__;
2140 
2141     int i, j, k, l;
2142     char buf[256];
2143     const CvHaarClassifierCascade* cascade = (const CvHaarClassifierCascade*) struct_ptr;
2144 
2145     /* TODO: parameters check */
2146 
2147     CV_CALL( cvStartWriteStruct( fs, name, CV_NODE_MAP, CV_TYPE_NAME_HAAR, attributes ) );
2148 
2149     CV_CALL( cvStartWriteStruct( fs, ICV_HAAR_SIZE_NAME, CV_NODE_SEQ | CV_NODE_FLOW ) );
2150     CV_CALL( cvWriteInt( fs, NULL, cascade->orig_window_size.width ) );
2151     CV_CALL( cvWriteInt( fs, NULL, cascade->orig_window_size.height ) );
2152     CV_CALL( cvEndWriteStruct( fs ) ); /* size */
2153 
2154     CV_CALL( cvStartWriteStruct( fs, ICV_HAAR_STAGES_NAME, CV_NODE_SEQ ) );
2155     for( i = 0; i < cascade->count; ++i )
2156     {
2157         CV_CALL( cvStartWriteStruct( fs, NULL, CV_NODE_MAP ) );
2158         sprintf( buf, "stage %d", i );
2159         CV_CALL( cvWriteComment( fs, buf, 1 ) );
2160 
2161         CV_CALL( cvStartWriteStruct( fs, ICV_HAAR_TREES_NAME, CV_NODE_SEQ ) );
2162 
2163         for( j = 0; j < cascade->stage_classifier[i].count; ++j )
2164         {
2165             CvHaarClassifier* tree = &cascade->stage_classifier[i].classifier[j];
2166 
2167             CV_CALL( cvStartWriteStruct( fs, NULL, CV_NODE_SEQ ) );
2168             sprintf( buf, "tree %d", j );
2169             CV_CALL( cvWriteComment( fs, buf, 1 ) );
2170 
2171             for( k = 0; k < tree->count; ++k )
2172             {
2173                 CvHaarFeature* feature = &tree->haar_feature[k];
2174 
2175                 CV_CALL( cvStartWriteStruct( fs, NULL, CV_NODE_MAP ) );
2176                 if( k )
2177                 {
2178                     sprintf( buf, "node %d", k );
2179                 }
2180                 else
2181                 {
2182                     sprintf( buf, "root node" );
2183                 }
2184                 CV_CALL( cvWriteComment( fs, buf, 1 ) );
2185 
2186                 CV_CALL( cvStartWriteStruct( fs, ICV_HAAR_FEATURE_NAME, CV_NODE_MAP ) );
2187 
2188                 CV_CALL( cvStartWriteStruct( fs, ICV_HAAR_RECTS_NAME, CV_NODE_SEQ ) );
2189                 for( l = 0; l < CV_HAAR_FEATURE_MAX && feature->rect[l].r.width != 0; ++l )
2190                 {
2191                     CV_CALL( cvStartWriteStruct( fs, NULL, CV_NODE_SEQ | CV_NODE_FLOW ) );
2192                     CV_CALL( cvWriteInt(  fs, NULL, feature->rect[l].r.x ) );
2193                     CV_CALL( cvWriteInt(  fs, NULL, feature->rect[l].r.y ) );
2194                     CV_CALL( cvWriteInt(  fs, NULL, feature->rect[l].r.width ) );
2195                     CV_CALL( cvWriteInt(  fs, NULL, feature->rect[l].r.height ) );
2196                     CV_CALL( cvWriteReal( fs, NULL, feature->rect[l].weight ) );
2197                     CV_CALL( cvEndWriteStruct( fs ) ); /* rect */
2198                 }
2199                 CV_CALL( cvEndWriteStruct( fs ) ); /* rects */
2200                 CV_CALL( cvWriteInt( fs, ICV_HAAR_TILTED_NAME, feature->tilted ) );
2201                 CV_CALL( cvEndWriteStruct( fs ) ); /* feature */
2202 
2203                 CV_CALL( cvWriteReal( fs, ICV_HAAR_THRESHOLD_NAME, tree->threshold[k]) );
2204 
2205                 if( tree->left[k] > 0 )
2206                 {
2207                     CV_CALL( cvWriteInt( fs, ICV_HAAR_LEFT_NODE_NAME, tree->left[k] ) );
2208                 }
2209                 else
2210                 {
2211                     CV_CALL( cvWriteReal( fs, ICV_HAAR_LEFT_VAL_NAME,
2212                         tree->alpha[-tree->left[k]] ) );
2213                 }
2214 
2215                 if( tree->right[k] > 0 )
2216                 {
2217                     CV_CALL( cvWriteInt( fs, ICV_HAAR_RIGHT_NODE_NAME, tree->right[k] ) );
2218                 }
2219                 else
2220                 {
2221                     CV_CALL( cvWriteReal( fs, ICV_HAAR_RIGHT_VAL_NAME,
2222                         tree->alpha[-tree->right[k]] ) );
2223                 }
2224 
2225                 CV_CALL( cvEndWriteStruct( fs ) ); /* split */
2226             }
2227 
2228             CV_CALL( cvEndWriteStruct( fs ) ); /* tree */
2229         }
2230 
2231         CV_CALL( cvEndWriteStruct( fs ) ); /* trees */
2232 
2233         CV_CALL( cvWriteReal( fs, ICV_HAAR_STAGE_THRESHOLD_NAME,
2234                               cascade->stage_classifier[i].threshold) );
2235 
2236         CV_CALL( cvWriteInt( fs, ICV_HAAR_PARENT_NAME,
2237                               cascade->stage_classifier[i].parent ) );
2238         CV_CALL( cvWriteInt( fs, ICV_HAAR_NEXT_NAME,
2239                               cascade->stage_classifier[i].next ) );
2240 
2241         CV_CALL( cvEndWriteStruct( fs ) ); /* stage */
2242     } /* for each stage */
2243 
2244     CV_CALL( cvEndWriteStruct( fs ) ); /* stages */
2245     CV_CALL( cvEndWriteStruct( fs ) ); /* root */
2246 
2247     __END__;
2248 }
2249 
2250 static void*
icvCloneHaarClassifier(const void * struct_ptr)2251 icvCloneHaarClassifier( const void* struct_ptr )
2252 {
2253     CvHaarClassifierCascade* cascade = NULL;
2254 
2255     CV_FUNCNAME( "cvCloneHaarClassifier" );
2256 
2257     __BEGIN__;
2258 
2259     int i, j, k, n;
2260     const CvHaarClassifierCascade* cascade_src =
2261         (const CvHaarClassifierCascade*) struct_ptr;
2262 
2263     n = cascade_src->count;
2264     CV_CALL( cascade = icvCreateHaarClassifierCascade(n) );
2265     cascade->orig_window_size = cascade_src->orig_window_size;
2266 
2267     for( i = 0; i < n; ++i )
2268     {
2269         cascade->stage_classifier[i].parent = cascade_src->stage_classifier[i].parent;
2270         cascade->stage_classifier[i].next = cascade_src->stage_classifier[i].next;
2271         cascade->stage_classifier[i].child = cascade_src->stage_classifier[i].child;
2272         cascade->stage_classifier[i].threshold = cascade_src->stage_classifier[i].threshold;
2273 
2274         cascade->stage_classifier[i].count = 0;
2275         CV_CALL( cascade->stage_classifier[i].classifier =
2276             (CvHaarClassifier*) cvAlloc( cascade_src->stage_classifier[i].count
2277                 * sizeof( cascade->stage_classifier[i].classifier[0] ) ) );
2278 
2279         cascade->stage_classifier[i].count = cascade_src->stage_classifier[i].count;
2280 
2281         for( j = 0; j < cascade->stage_classifier[i].count; ++j )
2282         {
2283             cascade->stage_classifier[i].classifier[j].haar_feature = NULL;
2284         }
2285 
2286         for( j = 0; j < cascade->stage_classifier[i].count; ++j )
2287         {
2288             const CvHaarClassifier* classifier_src =
2289                 &cascade_src->stage_classifier[i].classifier[j];
2290             CvHaarClassifier* classifier =
2291                 &cascade->stage_classifier[i].classifier[j];
2292 
2293             classifier->count = classifier_src->count;
2294             CV_CALL( classifier->haar_feature = (CvHaarFeature*) cvAlloc(
2295                 classifier->count * ( sizeof( *classifier->haar_feature ) +
2296                                       sizeof( *classifier->threshold ) +
2297                                       sizeof( *classifier->left ) +
2298                                       sizeof( *classifier->right ) ) +
2299                 (classifier->count + 1) * sizeof( *classifier->alpha ) ) );
2300             classifier->threshold = (float*) (classifier->haar_feature+classifier->count);
2301             classifier->left = (int*) (classifier->threshold + classifier->count);
2302             classifier->right = (int*) (classifier->left + classifier->count);
2303             classifier->alpha = (float*) (classifier->right + classifier->count);
2304             for( k = 0; k < classifier->count; ++k )
2305             {
2306                 classifier->haar_feature[k] = classifier_src->haar_feature[k];
2307                 classifier->threshold[k] = classifier_src->threshold[k];
2308                 classifier->left[k] = classifier_src->left[k];
2309                 classifier->right[k] = classifier_src->right[k];
2310                 classifier->alpha[k] = classifier_src->alpha[k];
2311             }
2312             classifier->alpha[classifier->count] =
2313                 classifier_src->alpha[classifier->count];
2314         }
2315     }
2316 
2317     __END__;
2318 
2319     return cascade;
2320 }
2321 
2322 
2323 CvType haar_type( CV_TYPE_NAME_HAAR, icvIsHaarClassifier,
2324                   (CvReleaseFunc)cvReleaseHaarClassifierCascade,
2325                   icvReadHaarClassifier, icvWriteHaarClassifier,
2326                   icvCloneHaarClassifier );
2327 
2328 /* End of file. */
2329