/
fhog.h
1404 lines (1233 loc) · 54.8 KB
/
fhog.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Copyright (C) 2013 Davis E. King (davis@dlib.net)
// License: Boost Software License See LICENSE.txt for the full license.
#ifndef DLIB_fHOG_Hh_
#define DLIB_fHOG_Hh_
#include "fhog_abstract.h"
#include "../matrix.h"
#include "../array2d.h"
#include "../array.h"
#include "../geometry.h"
#include "assign_image.h"
#include "draw.h"
#include "interpolation.h"
#include "../simd.h"
namespace dlib
{
// ----------------------------------------------------------------------------------------
namespace impl_fhog
{
template <typename image_type, typename T>
inline typename dlib::enable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient (
const int r,
const int c,
const image_type& img,
matrix<T,2,1>& grad,
T& len
)
{
matrix<T, 2, 1> grad2, grad3;
// get the red gradient
grad(0) = (int)img[r][c+1].red-(int)img[r][c-1].red;
grad(1) = (int)img[r+1][c].red-(int)img[r-1][c].red;
len = length_squared(grad);
// get the green gradient
grad2(0) = (int)img[r][c+1].green-(int)img[r][c-1].green;
grad2(1) = (int)img[r+1][c].green-(int)img[r-1][c].green;
T v2 = length_squared(grad2);
// get the blue gradient
grad3(0) = (int)img[r][c+1].blue-(int)img[r][c-1].blue;
grad3(1) = (int)img[r+1][c].blue-(int)img[r-1][c].blue;
T v3 = length_squared(grad3);
// pick color with strongest gradient
if (v2 > len)
{
len = v2;
grad = grad2;
}
if (v3 > len)
{
len = v3;
grad = grad3;
}
}
template <typename image_type>
inline typename dlib::enable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient (
const int r,
const int c,
const image_type& img,
simd4f& grad_x,
simd4f& grad_y,
simd4f& len
)
{
simd4i rleft((int)img[r][c-1].red,
(int)img[r][c].red,
(int)img[r][c+1].red,
(int)img[r][c+2].red);
simd4i rright((int)img[r][c+1].red,
(int)img[r][c+2].red,
(int)img[r][c+3].red,
(int)img[r][c+4].red);
simd4i rtop((int)img[r-1][c].red,
(int)img[r-1][c+1].red,
(int)img[r-1][c+2].red,
(int)img[r-1][c+3].red);
simd4i rbottom((int)img[r+1][c].red,
(int)img[r+1][c+1].red,
(int)img[r+1][c+2].red,
(int)img[r+1][c+3].red);
simd4i gleft((int)img[r][c-1].green,
(int)img[r][c].green,
(int)img[r][c+1].green,
(int)img[r][c+2].green);
simd4i gright((int)img[r][c+1].green,
(int)img[r][c+2].green,
(int)img[r][c+3].green,
(int)img[r][c+4].green);
simd4i gtop((int)img[r-1][c].green,
(int)img[r-1][c+1].green,
(int)img[r-1][c+2].green,
(int)img[r-1][c+3].green);
simd4i gbottom((int)img[r+1][c].green,
(int)img[r+1][c+1].green,
(int)img[r+1][c+2].green,
(int)img[r+1][c+3].green);
simd4i bleft((int)img[r][c-1].blue,
(int)img[r][c].blue,
(int)img[r][c+1].blue,
(int)img[r][c+2].blue);
simd4i bright((int)img[r][c+1].blue,
(int)img[r][c+2].blue,
(int)img[r][c+3].blue,
(int)img[r][c+4].blue);
simd4i btop((int)img[r-1][c].blue,
(int)img[r-1][c+1].blue,
(int)img[r-1][c+2].blue,
(int)img[r-1][c+3].blue);
simd4i bbottom((int)img[r+1][c].blue,
(int)img[r+1][c+1].blue,
(int)img[r+1][c+2].blue,
(int)img[r+1][c+3].blue);
simd4i grad_x_red = rright-rleft;
simd4i grad_y_red = rbottom-rtop;
simd4i grad_x_green = gright-gleft;
simd4i grad_y_green = gbottom-gtop;
simd4i grad_x_blue = bright-bleft;
simd4i grad_y_blue = bbottom-btop;
simd4i rlen = grad_x_red*grad_x_red + grad_y_red*grad_y_red;
simd4i glen = grad_x_green*grad_x_green + grad_y_green*grad_y_green;
simd4i blen = grad_x_blue*grad_x_blue + grad_y_blue*grad_y_blue;
simd4i cmp = rlen>glen;
simd4i tgrad_x = select(cmp,grad_x_red,grad_x_green);
simd4i tgrad_y = select(cmp,grad_y_red,grad_y_green);
simd4i tlen = select(cmp,rlen,glen);
cmp = tlen>blen;
grad_x = select(cmp,tgrad_x,grad_x_blue);
grad_y = select(cmp,tgrad_y,grad_y_blue);
len = select(cmp,tlen,blen);
}
// ------------------------------------------------------------------------------------
template <typename image_type>
inline typename dlib::enable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient(
const int r,
const int c,
const image_type& img,
simd8f& grad_x,
simd8f& grad_y,
simd8f& len
)
{
simd8i rleft((int)img[r][c - 1].red,
(int)img[r][c].red,
(int)img[r][c + 1].red,
(int)img[r][c + 2].red,
(int)img[r][c + 3].red,
(int)img[r][c + 4].red,
(int)img[r][c + 5].red,
(int)img[r][c + 6].red);
simd8i rright((int)img[r][c + 1].red,
(int)img[r][c + 2].red,
(int)img[r][c + 3].red,
(int)img[r][c + 4].red,
(int)img[r][c + 5].red,
(int)img[r][c + 6].red,
(int)img[r][c + 7].red,
(int)img[r][c + 8].red);
simd8i rtop((int)img[r - 1][c].red,
(int)img[r - 1][c + 1].red,
(int)img[r - 1][c + 2].red,
(int)img[r - 1][c + 3].red,
(int)img[r - 1][c + 4].red,
(int)img[r - 1][c + 5].red,
(int)img[r - 1][c + 6].red,
(int)img[r - 1][c + 7].red);
simd8i rbottom((int)img[r + 1][c].red,
(int)img[r + 1][c + 1].red,
(int)img[r + 1][c + 2].red,
(int)img[r + 1][c + 3].red,
(int)img[r + 1][c + 4].red,
(int)img[r + 1][c + 5].red,
(int)img[r + 1][c + 6].red,
(int)img[r + 1][c + 7].red);
simd8i gleft((int)img[r][c - 1].green,
(int)img[r][c].green,
(int)img[r][c + 1].green,
(int)img[r][c + 2].green,
(int)img[r][c + 3].green,
(int)img[r][c + 4].green,
(int)img[r][c + 5].green,
(int)img[r][c + 6].green);
simd8i gright((int)img[r][c + 1].green,
(int)img[r][c + 2].green,
(int)img[r][c + 3].green,
(int)img[r][c + 4].green,
(int)img[r][c + 5].green,
(int)img[r][c + 6].green,
(int)img[r][c + 7].green,
(int)img[r][c + 8].green);
simd8i gtop((int)img[r - 1][c].green,
(int)img[r - 1][c + 1].green,
(int)img[r - 1][c + 2].green,
(int)img[r - 1][c + 3].green,
(int)img[r - 1][c + 4].green,
(int)img[r - 1][c + 5].green,
(int)img[r - 1][c + 6].green,
(int)img[r - 1][c + 7].green);
simd8i gbottom((int)img[r + 1][c].green,
(int)img[r + 1][c + 1].green,
(int)img[r + 1][c + 2].green,
(int)img[r + 1][c + 3].green,
(int)img[r + 1][c + 4].green,
(int)img[r + 1][c + 5].green,
(int)img[r + 1][c + 6].green,
(int)img[r + 1][c + 7].green);
simd8i bleft((int)img[r][c - 1].blue,
(int)img[r][c].blue,
(int)img[r][c + 1].blue,
(int)img[r][c + 2].blue,
(int)img[r][c + 3].blue,
(int)img[r][c + 4].blue,
(int)img[r][c + 5].blue,
(int)img[r][c + 6].blue);
simd8i bright((int)img[r][c + 1].blue,
(int)img[r][c + 2].blue,
(int)img[r][c + 3].blue,
(int)img[r][c + 4].blue,
(int)img[r][c + 5].blue,
(int)img[r][c + 6].blue,
(int)img[r][c + 7].blue,
(int)img[r][c + 8].blue);
simd8i btop((int)img[r - 1][c].blue,
(int)img[r - 1][c + 1].blue,
(int)img[r - 1][c + 2].blue,
(int)img[r - 1][c + 3].blue,
(int)img[r - 1][c + 4].blue,
(int)img[r - 1][c + 5].blue,
(int)img[r - 1][c + 6].blue,
(int)img[r - 1][c + 7].blue);
simd8i bbottom((int)img[r + 1][c].blue,
(int)img[r + 1][c + 1].blue,
(int)img[r + 1][c + 2].blue,
(int)img[r + 1][c + 3].blue,
(int)img[r + 1][c + 4].blue,
(int)img[r + 1][c + 5].blue,
(int)img[r + 1][c + 6].blue,
(int)img[r + 1][c + 7].blue);
simd8i grad_x_red = rright - rleft;
simd8i grad_y_red = rbottom - rtop;
simd8i grad_x_green = gright - gleft;
simd8i grad_y_green = gbottom - gtop;
simd8i grad_x_blue = bright - bleft;
simd8i grad_y_blue = bbottom - btop;
simd8i rlen = grad_x_red*grad_x_red + grad_y_red*grad_y_red;
simd8i glen = grad_x_green*grad_x_green + grad_y_green*grad_y_green;
simd8i blen = grad_x_blue*grad_x_blue + grad_y_blue*grad_y_blue;
simd8i cmp = rlen > glen;
simd8i tgrad_x = select(cmp, grad_x_red, grad_x_green);
simd8i tgrad_y = select(cmp, grad_y_red, grad_y_green);
simd8i tlen = select(cmp, rlen, glen);
cmp = tlen > blen;
grad_x = select(cmp, tgrad_x, grad_x_blue);
grad_y = select(cmp, tgrad_y, grad_y_blue);
len = select(cmp, tlen, blen);
}
// ------------------------------------------------------------------------------------
template <typename image_type, typename T>
inline typename dlib::disable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient (
const int r,
const int c,
const image_type& img,
matrix<T, 2, 1>& grad,
T& len
)
{
grad(0) = (int)get_pixel_intensity(img[r][c+1])-(int)get_pixel_intensity(img[r][c-1]);
grad(1) = (int)get_pixel_intensity(img[r+1][c])-(int)get_pixel_intensity(img[r-1][c]);
len = length_squared(grad);
}
template <typename image_type>
inline typename dlib::disable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient (
int r,
int c,
const image_type& img,
simd4f& grad_x,
simd4f& grad_y,
simd4f& len
)
{
simd4i left((int)get_pixel_intensity(img[r][c-1]),
(int)get_pixel_intensity(img[r][c]),
(int)get_pixel_intensity(img[r][c+1]),
(int)get_pixel_intensity(img[r][c+2]));
simd4i right((int)get_pixel_intensity(img[r][c+1]),
(int)get_pixel_intensity(img[r][c+2]),
(int)get_pixel_intensity(img[r][c+3]),
(int)get_pixel_intensity(img[r][c+4]));
simd4i top((int)get_pixel_intensity(img[r-1][c]),
(int)get_pixel_intensity(img[r-1][c+1]),
(int)get_pixel_intensity(img[r-1][c+2]),
(int)get_pixel_intensity(img[r-1][c+3]));
simd4i bottom((int)get_pixel_intensity(img[r+1][c]),
(int)get_pixel_intensity(img[r+1][c+1]),
(int)get_pixel_intensity(img[r+1][c+2]),
(int)get_pixel_intensity(img[r+1][c+3]));
grad_x = right-left;
grad_y = bottom-top;
len = (grad_x*grad_x + grad_y*grad_y);
}
// ------------------------------------------------------------------------------------
template <typename image_type>
inline typename dlib::disable_if_c<pixel_traits<typename image_type::pixel_type>::rgb>::type get_gradient(
int r,
int c,
const image_type& img,
simd8f& grad_x,
simd8f& grad_y,
simd8f& len
)
{
simd8i left((int)get_pixel_intensity(img[r][c - 1]),
(int)get_pixel_intensity(img[r][c]),
(int)get_pixel_intensity(img[r][c + 1]),
(int)get_pixel_intensity(img[r][c + 2]),
(int)get_pixel_intensity(img[r][c + 3]),
(int)get_pixel_intensity(img[r][c + 4]),
(int)get_pixel_intensity(img[r][c + 5]),
(int)get_pixel_intensity(img[r][c + 6]));
simd8i right((int)get_pixel_intensity(img[r][c + 1]),
(int)get_pixel_intensity(img[r][c + 2]),
(int)get_pixel_intensity(img[r][c + 3]),
(int)get_pixel_intensity(img[r][c + 4]),
(int)get_pixel_intensity(img[r][c + 5]),
(int)get_pixel_intensity(img[r][c + 6]),
(int)get_pixel_intensity(img[r][c + 7]),
(int)get_pixel_intensity(img[r][c + 8]));
simd8i top((int)get_pixel_intensity(img[r - 1][c]),
(int)get_pixel_intensity(img[r - 1][c + 1]),
(int)get_pixel_intensity(img[r - 1][c + 2]),
(int)get_pixel_intensity(img[r - 1][c + 3]),
(int)get_pixel_intensity(img[r - 1][c + 4]),
(int)get_pixel_intensity(img[r - 1][c + 5]),
(int)get_pixel_intensity(img[r - 1][c + 6]),
(int)get_pixel_intensity(img[r - 1][c + 7]));
simd8i bottom((int)get_pixel_intensity(img[r + 1][c]),
(int)get_pixel_intensity(img[r + 1][c + 1]),
(int)get_pixel_intensity(img[r + 1][c + 2]),
(int)get_pixel_intensity(img[r + 1][c + 3]),
(int)get_pixel_intensity(img[r + 1][c + 4]),
(int)get_pixel_intensity(img[r + 1][c + 5]),
(int)get_pixel_intensity(img[r + 1][c + 6]),
(int)get_pixel_intensity(img[r + 1][c + 7]));
grad_x = right - left;
grad_y = bottom - top;
len = (grad_x*grad_x + grad_y*grad_y);
}
// ------------------------------------------------------------------------------------
template <typename T, typename mm1, typename mm2>
inline void set_hog (
dlib::array<array2d<T,mm1>,mm2>& hog,
int o,
int x,
int y,
const float& value
)
{
hog[o][y][x] = value;
}
template <typename T, typename mm1, typename mm2>
void init_hog (
dlib::array<array2d<T,mm1>,mm2>& hog,
int hog_nr,
int hog_nc,
int filter_rows_padding,
int filter_cols_padding
)
{
const int num_hog_bands = 27+4;
hog.resize(num_hog_bands);
for (int i = 0; i < num_hog_bands; ++i)
{
hog[i].set_size(hog_nr+filter_rows_padding-1, hog_nc+filter_cols_padding-1);
rectangle rect = get_rect(hog[i]);
rect.top() += (filter_rows_padding-1)/2;
rect.left() += (filter_cols_padding-1)/2;
rect.right() -= filter_cols_padding/2;
rect.bottom() -= filter_rows_padding/2;
zero_border_pixels(hog[i],rect);
}
}
template <typename T, typename mm1, typename mm2>
void init_hog_zero_everything (
dlib::array<array2d<T,mm1>,mm2>& hog,
int hog_nr,
int hog_nc,
int filter_rows_padding,
int filter_cols_padding
)
{
const int num_hog_bands = 27+4;
hog.resize(num_hog_bands);
for (int i = 0; i < num_hog_bands; ++i)
{
hog[i].set_size(hog_nr+filter_rows_padding-1, hog_nc+filter_cols_padding-1);
assign_all_pixels(hog[i], 0);
}
}
// ------------------------------------------------------------------------------------
template <typename T, typename mm>
inline void set_hog (
array2d<matrix<T,31,1>,mm>& hog,
int o,
int x,
int y,
const float& value
)
{
hog[y][x](o) = value;
}
template <typename T, typename mm>
void init_hog (
array2d<matrix<T,31,1>,mm>& hog,
int hog_nr,
int hog_nc,
int filter_rows_padding,
int filter_cols_padding
)
{
hog.set_size(hog_nr+filter_rows_padding-1, hog_nc+filter_cols_padding-1);
// now zero out the border region
rectangle rect = get_rect(hog);
rect.top() += (filter_rows_padding-1)/2;
rect.left() += (filter_cols_padding-1)/2;
rect.right() -= filter_cols_padding/2;
rect.bottom() -= filter_rows_padding/2;
border_enumerator be(get_rect(hog),rect);
while (be.move_next())
{
const point p = be.element();
set_all_elements(hog[p.y()][p.x()], 0);
}
}
template <typename T, typename mm>
void init_hog_zero_everything (
array2d<matrix<T,31,1>,mm>& hog,
int hog_nr,
int hog_nc,
int filter_rows_padding,
int filter_cols_padding
)
{
hog.set_size(hog_nr+filter_rows_padding-1, hog_nc+filter_cols_padding-1);
for (long r = 0; r < hog.nr(); ++r)
{
for (long c = 0; c < hog.nc(); ++c)
{
set_all_elements(hog[r][c], 0);
}
}
}
// ------------------------------------------------------------------------------------
template <
typename image_type,
typename out_type
>
void impl_extract_fhog_features_cell_size_1(
const image_type& img_,
out_type& hog,
int filter_rows_padding,
int filter_cols_padding
)
{
const_image_view<image_type> img(img_);
// make sure requires clause is not broken
DLIB_ASSERT( filter_rows_padding > 0 &&
filter_cols_padding > 0 ,
"\t void extract_fhog_features()"
<< "\n\t Invalid inputs were given to this function. "
<< "\n\t filter_rows_padding: " << filter_rows_padding
<< "\n\t filter_cols_padding: " << filter_cols_padding
);
/*
This function is an optimized version of impl_extract_fhog_features() for
the case where cell_size == 1.
*/
// unit vectors used to compute gradient orientation
matrix<float,2,1> directions[9];
directions[0] = 1.0000, 0.0000;
directions[1] = 0.9397, 0.3420;
directions[2] = 0.7660, 0.6428;
directions[3] = 0.500, 0.8660;
directions[4] = 0.1736, 0.9848;
directions[5] = -0.1736, 0.9848;
directions[6] = -0.5000, 0.8660;
directions[7] = -0.7660, 0.6428;
directions[8] = -0.9397, 0.3420;
if (img.nr() <= 2 || img.nc() <= 2)
{
hog.clear();
return;
}
array2d<unsigned char> angle(img.nr(), img.nc());
array2d<float> norm(img.nr(), img.nc());
zero_border_pixels(norm,1,1);
// memory for HOG features
const long hog_nr = img.nr()-2;
const long hog_nc = img.nc()-2;
const int padding_rows_offset = (filter_rows_padding-1)/2;
const int padding_cols_offset = (filter_cols_padding-1)/2;
init_hog_zero_everything(hog, hog_nr, hog_nc, filter_rows_padding, filter_cols_padding);
const int visible_nr = img.nr()-1;
const int visible_nc = img.nc()-1;
// First populate the gradient histograms
for (int y = 1; y < visible_nr; y++)
{
int x;
for (x = 1; x < visible_nc - 7; x += 8)
{
// v will be the length of the gradient vectors.
simd8f grad_x, grad_y, v;
get_gradient(y, x, img, grad_x, grad_y, v);
float _vv[8];
v.store(_vv);
// Now snap the gradient to one of 18 orientations
simd8f best_dot = 0;
simd8f best_o = 0;
for (int o = 0; o < 9; o++)
{
simd8f dot = grad_x*directions[o](0) + grad_y*directions[o](1);
simd8f_bool cmp = dot>best_dot;
best_dot = select(cmp, dot, best_dot);
dot *= -1;
best_o = select(cmp, o, best_o);
cmp = dot > best_dot;
best_dot = select(cmp, dot, best_dot);
best_o = select(cmp, o + 9, best_o);
}
int32 _best_o[8]; simd8i(best_o).store(_best_o);
norm[y][x + 0] = _vv[0];
norm[y][x + 1] = _vv[1];
norm[y][x + 2] = _vv[2];
norm[y][x + 3] = _vv[3];
norm[y][x + 4] = _vv[4];
norm[y][x + 5] = _vv[5];
norm[y][x + 6] = _vv[6];
norm[y][x + 7] = _vv[7];
angle[y][x + 0] = _best_o[0];
angle[y][x + 1] = _best_o[1];
angle[y][x + 2] = _best_o[2];
angle[y][x + 3] = _best_o[3];
angle[y][x + 4] = _best_o[4];
angle[y][x + 5] = _best_o[5];
angle[y][x + 6] = _best_o[6];
angle[y][x + 7] = _best_o[7];
}
// Now process the right columns that don't fit into simd registers.
for (; x < visible_nc; x++)
{
matrix<float,2,1> grad;
float v;
get_gradient(y,x,img,grad,v);
// snap to one of 18 orientations
float best_dot = 0;
int best_o = 0;
for (int o = 0; o < 9; o++)
{
const float dot = dlib::dot(directions[o], grad);
if (dot > best_dot)
{
best_dot = dot;
best_o = o;
}
else if (-dot > best_dot)
{
best_dot = -dot;
best_o = o+9;
}
}
norm[y][x] = v;
angle[y][x] = best_o;
}
}
const float eps = 0.0001;
// compute features
for (int y = 0; y < hog_nr; y++)
{
const int yy = y+padding_rows_offset;
for (int x = 0; x < hog_nc; x++)
{
const simd4f z1(norm[y+1][x+1],
norm[y][x+1],
norm[y+1][x],
norm[y][x]);
const simd4f z2(norm[y+1][x+2],
norm[y][x+2],
norm[y+1][x+1],
norm[y][x+1]);
const simd4f z3(norm[y+2][x+1],
norm[y+1][x+1],
norm[y+2][x],
norm[y+1][x]);
const simd4f z4(norm[y+2][x+2],
norm[y+1][x+2],
norm[y+2][x+1],
norm[y+1][x+1]);
const simd4f temp0 = std::sqrt(norm[y+1][x+1]);
const simd4f nn = 0.2*sqrt(z1+z2+z3+z4+eps);
const simd4f n = 0.1/nn;
simd4f t = 0;
const int xx = x+padding_cols_offset;
simd4f h0 = min(temp0,nn)*n;
const float vv = sum(h0);
set_hog(hog,angle[y+1][x+1],xx,yy, vv);
t += h0;
t *= 2*0.2357;
// contrast-insensitive features
set_hog(hog,angle[y+1][x+1]%9+18,xx,yy, vv);
float temp[4];
t.store(temp);
// texture features
set_hog(hog,27,xx,yy, temp[0]);
set_hog(hog,28,xx,yy, temp[1]);
set_hog(hog,29,xx,yy, temp[2]);
set_hog(hog,30,xx,yy, temp[3]);
}
}
}
// ------------------------------------------------------------------------------------
template <
typename image_type,
typename out_type
>
void impl_extract_fhog_features(
const image_type& img_,
out_type& hog,
int cell_size,
int filter_rows_padding,
int filter_cols_padding
)
{
const_image_view<image_type> img(img_);
// make sure requires clause is not broken
DLIB_ASSERT( cell_size > 0 &&
filter_rows_padding > 0 &&
filter_cols_padding > 0 ,
"\t void extract_fhog_features()"
<< "\n\t Invalid inputs were given to this function. "
<< "\n\t cell_size: " << cell_size
<< "\n\t filter_rows_padding: " << filter_rows_padding
<< "\n\t filter_cols_padding: " << filter_cols_padding
);
/*
This function implements the HOG feature extraction method described in
the paper:
P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan
Object Detection with Discriminatively Trained Part Based Models
IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 32, No. 9, Sep. 2010
Moreover, this function is derived from the HOG feature extraction code
from the features.cc file in the voc-releaseX code (see
http://people.cs.uchicago.edu/~rbg/latent/) which is has the following
license (note that the code has been modified to work with grayscale and
color as well as planar and interlaced input and output formats):
Copyright (C) 2011, 2012 Ross Girshick, Pedro Felzenszwalb
Copyright (C) 2008, 2009, 2010 Pedro Felzenszwalb, Ross Girshick
Copyright (C) 2007 Pedro Felzenszwalb, Deva Ramanan
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
if (cell_size == 1)
{
impl_extract_fhog_features_cell_size_1(img_,hog,filter_rows_padding,filter_cols_padding);
return;
}
// unit vectors used to compute gradient orientation
matrix<float,2,1> directions[9];
directions[0] = 1.0000, 0.0000;
directions[1] = 0.9397, 0.3420;
directions[2] = 0.7660, 0.6428;
directions[3] = 0.500, 0.8660;
directions[4] = 0.1736, 0.9848;
directions[5] = -0.1736, 0.9848;
directions[6] = -0.5000, 0.8660;
directions[7] = -0.7660, 0.6428;
directions[8] = -0.9397, 0.3420;
// First we allocate memory for caching orientation histograms & their norms.
const int cells_nr = (int)((float)img.nr()/(float)cell_size + 0.5);
const int cells_nc = (int)((float)img.nc()/(float)cell_size + 0.5);
if (cells_nr == 0 || cells_nc == 0)
{
hog.clear();
return;
}
// We give hist extra padding around the edges (1 cell all the way around the
// edge) so we can avoid needing to do boundary checks when indexing into it
// later on. So some statements assign to the boundary but those values are
// never used.
array2d<matrix<float,18,1> > hist(cells_nr+2, cells_nc+2);
for (long r = 0; r < hist.nr(); ++r)
{
for (long c = 0; c < hist.nc(); ++c)
{
hist[r][c] = 0;
}
}
array2d<float> norm(cells_nr, cells_nc);
assign_all_pixels(norm, 0);
// memory for HOG features
const int hog_nr = std::max(cells_nr-2, 0);
const int hog_nc = std::max(cells_nc-2, 0);
if (hog_nr == 0 || hog_nc == 0)
{
hog.clear();
return;
}
const int padding_rows_offset = (filter_rows_padding-1)/2;
const int padding_cols_offset = (filter_cols_padding-1)/2;
init_hog(hog, hog_nr, hog_nc, filter_rows_padding, filter_cols_padding);
const int visible_nr = std::min(cells_nr*cell_size,static_cast<int>(img.nr()))-1;
const int visible_nc = std::min(cells_nc*cell_size,static_cast<int>(img.nc()))-1;
// First populate the gradient histograms
for (int y = 1; y < visible_nr; y++)
{
const float yp = (y + 0.5) / static_cast<float>(cell_size) - 0.5;
const int iyp = static_cast<int>(std::floor(yp));
const float vy0 = yp - iyp;
const float vy1 = 1.0 - vy0;
int x;
for (x = 1; x < visible_nc - 7; x += 8)
{
simd8f xx(x, x + 1, x + 2, x + 3, x + 4, x + 5, x + 6, x + 7);
// v will be the length of the gradient vectors.
simd8f grad_x, grad_y, v;
get_gradient(y, x, img, grad_x, grad_y, v);
// We will use bilinear interpolation to add into the histogram bins.
// So first we precompute the values needed to determine how much each
// pixel votes into each bin.
simd8f xp = (xx + 0.5) / static_cast<float>(cell_size) + 0.5;
simd8i ixp = simd8i(xp);
simd8f vx0 = xp - ixp;
simd8f vx1 = 1.0f - vx0;
v = sqrt(v);
// Now snap the gradient to one of 18 orientations
simd8f best_dot = 0;
simd8f best_o = 0;
for (int o = 0; o < 9; o++)
{
simd8f dot = grad_x*directions[o](0) + grad_y*directions[o](1);
simd8f_bool cmp = dot>best_dot;
best_dot = select(cmp, dot, best_dot);
dot *= -1;
best_o = select(cmp, o, best_o);
cmp = dot > best_dot;
best_dot = select(cmp, dot, best_dot);
best_o = select(cmp, o + 9, best_o);
}
// Add the gradient magnitude, v, to 4 histograms around pixel using
// bilinear interpolation.
vx1 *= v;
vx0 *= v;
// The amounts for each bin
simd8f v11 = vy1*vx1;
simd8f v01 = vy0*vx1;
simd8f v10 = vy1*vx0;
simd8f v00 = vy0*vx0;
int32 _best_o[8]; simd8i(best_o).store(_best_o);
int32 _ixp[8]; ixp.store(_ixp);
float _v11[8]; v11.store(_v11);
float _v01[8]; v01.store(_v01);
float _v10[8]; v10.store(_v10);
float _v00[8]; v00.store(_v00);
hist[iyp + 1][_ixp[0]](_best_o[0]) += _v11[0];
hist[iyp + 1 + 1][_ixp[0]](_best_o[0]) += _v01[0];
hist[iyp + 1][_ixp[0] + 1](_best_o[0]) += _v10[0];
hist[iyp + 1 + 1][_ixp[0] + 1](_best_o[0]) += _v00[0];
hist[iyp + 1][_ixp[1]](_best_o[1]) += _v11[1];
hist[iyp + 1 + 1][_ixp[1]](_best_o[1]) += _v01[1];
hist[iyp + 1][_ixp[1] + 1](_best_o[1]) += _v10[1];
hist[iyp + 1 + 1][_ixp[1] + 1](_best_o[1]) += _v00[1];
hist[iyp + 1][_ixp[2]](_best_o[2]) += _v11[2];
hist[iyp + 1 + 1][_ixp[2]](_best_o[2]) += _v01[2];
hist[iyp + 1][_ixp[2] + 1](_best_o[2]) += _v10[2];
hist[iyp + 1 + 1][_ixp[2] + 1](_best_o[2]) += _v00[2];
hist[iyp + 1][_ixp[3]](_best_o[3]) += _v11[3];
hist[iyp + 1 + 1][_ixp[3]](_best_o[3]) += _v01[3];
hist[iyp + 1][_ixp[3] + 1](_best_o[3]) += _v10[3];
hist[iyp + 1 + 1][_ixp[3] + 1](_best_o[3]) += _v00[3];
hist[iyp + 1][_ixp[4]](_best_o[4]) += _v11[4];
hist[iyp + 1 + 1][_ixp[4]](_best_o[4]) += _v01[4];
hist[iyp + 1][_ixp[4] + 1](_best_o[4]) += _v10[4];
hist[iyp + 1 + 1][_ixp[4] + 1](_best_o[4]) += _v00[4];
hist[iyp + 1][_ixp[5]](_best_o[5]) += _v11[5];
hist[iyp + 1 + 1][_ixp[5]](_best_o[5]) += _v01[5];
hist[iyp + 1][_ixp[5] + 1](_best_o[5]) += _v10[5];
hist[iyp + 1 + 1][_ixp[5] + 1](_best_o[5]) += _v00[5];
hist[iyp + 1][_ixp[6]](_best_o[6]) += _v11[6];
hist[iyp + 1 + 1][_ixp[6]](_best_o[6]) += _v01[6];
hist[iyp + 1][_ixp[6] + 1](_best_o[6]) += _v10[6];
hist[iyp + 1 + 1][_ixp[6] + 1](_best_o[6]) += _v00[6];
hist[iyp + 1][_ixp[7]](_best_o[7]) += _v11[7];
hist[iyp + 1 + 1][_ixp[7]](_best_o[7]) += _v01[7];
hist[iyp + 1][_ixp[7] + 1](_best_o[7]) += _v10[7];
hist[iyp + 1 + 1][_ixp[7] + 1](_best_o[7]) += _v00[7];
}
// Now process the right columns that don't fit into simd registers.
for (; x < visible_nc; x++)
{
matrix<float, 2, 1> grad;
float v;
get_gradient(y,x,img,grad,v);
// snap to one of 18 orientations
float best_dot = 0;
int best_o = 0;
for (int o = 0; o < 9; o++)
{
const float dot = dlib::dot(directions[o], grad);
if (dot > best_dot)
{
best_dot = dot;
best_o = o;
}
else if (-dot > best_dot)
{
best_dot = -dot;
best_o = o+9;
}
}
v = std::sqrt(v);
// add to 4 histograms around pixel using bilinear interpolation
const float xp = (x + 0.5) / static_cast<double>(cell_size) - 0.5;
const int ixp = static_cast<int>(std::floor(xp));
const float vx0 = xp - ixp;
const float vx1 = 1.0 - vx0;
hist[iyp+1][ixp+1](best_o) += vy1*vx1*v;
hist[iyp+1+1][ixp+1](best_o) += vy0*vx1*v;
hist[iyp+1][ixp+1+1](best_o) += vy1*vx0*v;
hist[iyp+1+1][ixp+1+1](best_o) += vy0*vx0*v;
}
}
// compute energy in each block by summing over orientations
for (int r = 0; r < cells_nr; ++r)
{
for (int c = 0; c < cells_nc; ++c)
{
for (int o = 0; o < 9; o++)
{
norm[r][c] += (hist[r+1][c+1](o) + hist[r+1][c+1](o+9)) * (hist[r+1][c+1](o) + hist[r+1][c+1](o+9));
}
}
}
const float eps = 0.0001;
// compute features
for (int y = 0; y < hog_nr; y++)
{
const int yy = y+padding_rows_offset;
for (int x = 0; x < hog_nc; x++)
{
const simd4f z1(norm[y+1][x+1],
norm[y][x+1],
norm[y+1][x],
norm[y][x]);
const simd4f z2(norm[y+1][x+2],
norm[y][x+2],
norm[y+1][x+1],
norm[y][x+1]);
const simd4f z3(norm[y+2][x+1],
norm[y+1][x+1],
norm[y+2][x],
norm[y+1][x]);
const simd4f z4(norm[y+2][x+2],
norm[y+1][x+2],
norm[y+2][x+1],
norm[y+1][x+1]);
const simd4f nn = 0.2*sqrt(z1+z2+z3+z4+eps);
const simd4f n = 0.1/nn;
simd4f t = 0;