515:(HOG), gradient histogram channel (Hist), gradient magnitude (Grad), color channels (RGB, HSV, LUV) and grayscale channel. The performance was evaluated in terms of pedestrian detection rates at the reference point of 10 - 4 fppw (false positive per window). HOG turned out to be the most informative channel compared with rest of the channels. The detection rate of HOG was 89%. Further, among the color channels (RGB, HSV and LUV), LUV had the best detection rate of 55.8%.
102:. According to this notion of a channel, color channels of an image can be redefined as output images that are obtained by extracting one specific color information point from the input image at a time. Similarly, a channel for a grayscale input image is simply equal to a grayscale input image. The simple MATLAB implementation below shows how color channels and grayscale channel can be extracted from an input image.
544:.. In fact, it is now used as a baseline detector due to its proven efficiency and reasonable performance. Several authors have obtained even better performance by either extending feature pool in various ways or by carefully choosing the classifier and training it with a larger dataset. Work by Zhang
519:
channel was least informative with the detection rate of only 30.7%. Next, they evaluated the performance of various channel combinations, which is their proposed method. The combination of LUV, Hist and Grad channels had the highest detection rate of 91.9%. This channel combination was further used
498:
The ChnFtrs method allows one to pool features that capture the richness from diverse channels. Dollár, et al. based their experimental results on first order features since there was not much added value by the second order features. The channels are re-computed at multiple scales to extract a pool
548:
also exploited integral channel features in developing
Informed Haar detector for pedestrian detection. They used the same combination of channels as Dollár et al. but were able achieve approximately 20% higher performance than the baseline ChnFtrs method. The added performance was due to the fact
531:
detector was tested on full images from INRIA and
Caltech datasets. The performance was compared with 12 other detectors including HOG, which is the most popular method. ChnFtrs outperformed all except LatSvm. The detection rate for ChnFtrs was 86% on INRIA dataset and 60% on a more challenging
94:. A color image, for example is an aggregate of three channels (red, green and blue). The color data of an image is stored in three arrays of values, known as channels. While this definition of a "channel" is widely accepted across various domains, there exists a broader definition in
454:
Pointwise transformations: As the name suggests, pointwise transformations operate on individual pixels. The examples include logarithmic operator and exponential operator. The logarithmic operator enhances low intensity pixels whereas exponential does the complete
229:
It is clear from the above examples that a channel can be generated by either simply extracting specific information from the original image or by manipulating the input image in some form to obtain the desired channel. Dollár
549:
that they provided better prior knowledge to their detector. It is also important to note that they used informed Haar-like features, which are second order features according to the terminology described in, whereas Dollár
560:
ChnFtrs is a versatile method that allows one to extract features from multiple channels, thus allowing to capture diverse information from a single input image. The performance of a base detector developed by Dollár
499:
of channel features that can represent the entire scale space. There is a MATLAB toolbox that can be used as a guidance to implement ChnFtrs method. Further, OpenCV has a complete implementation of ChnFtrs.
486:). Note that it is computationally less expensive to extract local sums from an integral image rather than from image pixels. In fact, Dollár, et al., used integral images in their work to extract features.
632:
S. Zhang, C. Bauckhage, and A. Cremers. Informed Haar-like features improve pedestrian detection. In
Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on, pages 947–954. IEEE, 2014
553:. demonstrated their results using first order channel features only, as their analysis showed that second order features barely added 0.6% increase to their detection rate. Further, Benenson
474:
Once channels are obtained from an input image, various features can be extracted from these channels. These features are called channel features and can be categorized into two main types:
270:
Linear filters: This is a simple method for generating channels. There are variety of linear filters that allow us to capture different aspects of an image. A few examples are
458:
Gradient histogram: It is a histogram of an image where bins are determined by the gradient angle. Each pixel votes and the weight is determined by its gradient magnitude.
63:
Extract features such as sums over rectangular channel regions from each channel. The features extracted from various channels are called integral channel features.
234:. defined a channel generation function as Ω, which can be used to relate a channel (that is, an output image) to the original image as follows.
263:
Color and grayscale: As discussed above, we can easily extract color and grayscale channels from an image. Note that color channels could be
478:
First-order channel features: Features extracted from a single channel by summing pixels in fixed rectangular regions. These are denoted as
657:
540:
The ICF method (ChnFtrs) has been widely exploited by researchers in
Computer Vision after the work was initially published by Dollar
98:, which allows one to exploit other features of an image besides the color information. One such definition refers to a channel as a
489:
Higher-order channel features: Features obtained by combining two or more first order channel features. For example, Haar features.
74:
used boosting technique which offers faster learning but training could be done with any of the other available methods such as
100:
registered map of the original image where the output pixels are mapped to input pixels by some linear or non-transformation
616:
282:
and DoG capture edge information and textured-ness of an image. Below is a sample code for implementing DoG in MATLAB.
557:. were able to increase the detection speed of baseline ChnFtrs method by avoiding the need to resize input image.
641:
R. Benenson, M. Mathias, R. Timofte, and L. V. Gool. Pedestrian detection at 100 frames per second. In CVPR, 2012
512:
459:
250:
The next section discusses other relatively complex channel types as mentioned in the original paper by Dollár
60:
Compute multiple registered image channels from an input image, using linear and non-linear transformations
511:. first evaluated the effectiveness of various channels when used individually. The channels studied were
451:
and gradient magnitude. Canny gives the edge information whereas gradient magnitude gives edge strength.
565:. has been shown to be enhanced by adding better prior knowledge and training with a larger dataset.
275:
75:
48:
267:, LUV or HSV. LUV color channels have shown to be most informative among all color spaces.
8:
448:
40:
90:
Typically, a "channel" refers to a certain component that defines pixel values in a
43:
from multiple registered image channels. This method was highly exploited by Dollár
28:
592:
P. Dollár, Z. Tu, P. Perona and S. Belongie "Integral
Channel Features", BMVC 2009
447:
Nonlinear channels: There are many non-linear channels. The most popular ones are
95:
36:
32:
462:(HOG) descriptor is a popular descriptor that was developed by Dalal and Triggs.
602:
523:
About 30,000 first order features were used to train AdaBoost classifier. The
466:
Note that these channels can be used alone or in combination with each other.
651:
91:
279:
271:
131:% where color channel could be red, green or blue. The three output images
203:% Note if input image I was already a grayscale image, grayscale channel
617:"Integral Channel Features Detector — OpenCV 3.0.0-dev documentation"
516:
206:% would have simply been equal to input image, i.e., gray channel = I
528:
67:
254:. MATLAB implementation is given for some of the channels.
493:
264:
39:
to extract features such as local sums, histograms and
81:
Finally, trained classifier is used to detect objects
520:
in their experiments on INRIA and
Caltech datasets.
649:
51:, that was first described at the BMVC in 2009.
291:% Difference of Gaussian applied on input image
595:
635:
628:
626:
507:To study the performance of ChnFtrs, Dollár
134:% are extracted from input image as follows
588:
586:
584:
582:
580:
578:
623:
609:
54:
257:
575:
494:Integral channel feature implementation
650:
535:
85:
469:
200:% Output image = grayscale_image(I).
658:Feature detection (computer vision)
324:% create a Gaussian with signal 0.5
13:
128:% Output_image = color_channel(I),
14:
669:
357:% create a Gaussian with signal 3
513:histogram of oriented gradients
460:Histogram of oriented gradients
441:% convolve DoG with input image
502:
1:
568:
7:
10:
674:
17:Integral Channel Features
603:"Piotr's Matlab Toolbox"
285:
104:
288:% Output image = DoG(I)
276:difference of Gaussians
76:support vector machine
55:Overview of the method
258:Various channel types
49:pedestrian detection
47:. in their work for
536:Further development
449:canny edge detector
405:'RGB_1.jpg'
242:is the channel and
238:𝐶 = Ω(𝐼) , where
125:% input color image
119:'I_RGB.png'
86:Images and channels
70:classifier. Dollár
470:Feature extraction
339:'gaussian'
306:'gaussian'
41:Haar-like features
27:, is a method for
532:Caltech dataset.
246:is an input image
23:), also known as
665:
642:
639:
633:
630:
621:
620:
613:
607:
606:
599:
593:
590:
442:
439:
436:
433:
430:
427:
424:
421:
418:
415:
412:
409:
406:
403:
400:
397:
394:
391:
388:
385:
382:
379:
376:
373:
370:
367:
364:
361:
358:
355:
352:
349:
346:
343:
340:
337:
334:
331:
328:
325:
322:
319:
316:
313:
310:
307:
304:
301:
298:
295:
292:
289:
225:
222:
219:
216:
213:
210:
207:
204:
201:
198:
195:
192:
189:
186:
183:
180:
177:
174:
171:
168:
165:
162:
159:
156:
153:
150:
147:
144:
141:
138:
135:
132:
129:
126:
123:
120:
117:
114:
111:
108:
29:object detection
673:
672:
668:
667:
666:
664:
663:
662:
648:
647:
646:
645:
640:
636:
631:
624:
615:
614:
610:
601:
600:
596:
591:
576:
571:
538:
505:
496:
472:
444:
443:
440:
437:
434:
431:
428:
425:
422:
419:
416:
413:
410:
407:
404:
401:
398:
395:
392:
389:
386:
383:
380:
377:
374:
371:
368:
365:
362:
359:
356:
353:
350:
347:
344:
341:
338:
335:
332:
329:
326:
323:
320:
317:
314:
311:
308:
305:
302:
299:
296:
293:
290:
287:
260:
227:
226:
223:
220:
217:
214:
211:
208:
205:
202:
199:
196:
193:
190:
187:
184:
181:
178:
175:
172:
169:
166:
163:
160:
157:
154:
151:
148:
145:
142:
139:
136:
133:
130:
127:
124:
121:
118:
115:
112:
109:
106:
96:computer vision
88:
57:
37:integral images
33:computer vision
12:
11:
5:
671:
661:
660:
644:
643:
634:
622:
608:
594:
573:
572:
570:
567:
537:
534:
504:
501:
495:
492:
491:
490:
487:
471:
468:
464:
463:
456:
452:
435:'same'
378:% create a DoG
286:
284:
283:
268:
259:
256:
248:
247:
105:
87:
84:
83:
82:
79:
64:
61:
56:
53:
9:
6:
4:
3:
2:
670:
659:
656:
655:
653:
638:
629:
627:
618:
612:
604:
598:
589:
587:
585:
583:
581:
579:
574:
566:
564:
558:
556:
552:
547:
543:
533:
530:
526:
521:
518:
514:
510:
500:
488:
485:
481:
477:
476:
475:
467:
461:
457:
453:
450:
446:
445:
281:
277:
273:
269:
266:
262:
261:
255:
253:
245:
241:
237:
236:
235:
233:
158:green_channel
103:
101:
97:
93:
92:digital image
80:
77:
73:
69:
65:
62:
59:
58:
52:
50:
46:
42:
38:
34:
30:
26:
22:
18:
637:
611:
597:
562:
559:
554:
550:
545:
541:
539:
524:
522:
508:
506:
497:
483:
479:
473:
465:
280:Gabor filter
272:Gabor filter
251:
249:
243:
239:
231:
228:
209:gray_channel
179:blue_channel
99:
89:
71:
44:
24:
20:
16:
15:
503:Performance
411:DoG_channel
137:red_channel
569:References
429:DoG_filter
360:DoG_filter
66:Train the
35:. It uses
517:Grayscale
455:opposite.
652:Category
529:AdaBoost
393:rgb2gray
333:fspecial
300:fspecial
215:rgb2gray
68:AdaBoost
525:ChnFtrs
278:(DoG).
25:ChnFtrs
399:imread
387:double
113:imread
72:et al.
563:et al
555:et al
551:et al
546:et al
542:et al
509:et al
423:image
417:conv2
381:image
252:et al
232:et al
45:et al
408:)));
274:and
318:0.5
265:RGB
188:(:,
167:(:,
146:(:,
31:in
21:ICF
654::
625:^
577:^
527:+
438:);
372:H2
366:H1
354:);
345:25
327:H2
321:);
312:25
294:H1
224:);
197:);
191::,
176:);
170::,
155:);
149::,
122:);
619:.
605:.
484:C
482:(
480:ƒ
432:,
426:,
420:(
414:=
402:(
396:(
390:(
384:=
375:;
369:-
363:=
351:3
348:,
342:,
336:(
330:=
315:,
309:,
303:(
297:=
244:I
240:C
221:I
218:(
212:=
194:3
185:I
182:=
173:2
164:I
161:=
152:1
143:I
140:=
116:(
110:=
107:I
78:.
19:(
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.