The method of localization of persons using the method of fast comparison based on the OSAD algorithm (excerpt from the publication)
Habrovchane, hello!
As you probably know, Innopolis University is starting a series of webinars with our teachers. The first webinar on Artificial Intelligence will be held on February 11, 2014 at 18:00 Moscow time. He will be held by Associate Professor of Innopolis University Samir Belhauari .
Registration link - attendee.gotowebinar.com/register/6601261461187578113 . Hurry up to register today, the number of places is limited!
')
Partial translation of the article on the topic of facial recognition, published by S. Belkhauari in the International Journal of Computer Applications, read under the cut.
Professor Belhauari sends you his greeting:
The article was published in International Journal of Computer Applications (0975 - 8887), Volume 18– No.8, March 2011.
The method of localization of persons using the method of quick comparison with the standard based on the algorithm optimized sum of absolute differences, N. Dawoud, S. Belhaouari, J.Janier
annotation
Until recently, the method of recognition by comparing patterns (standards) was widely used to solve the problems of localization of faces in images. The normalized cross-correlation function (NEC) is considered a measurement method traditionally used to calculate the similarity between existing face patterns and rectangular blocks of an input image to localize the position of a face in a picture. However, there is always an error in determining the position of a face due to the fact that some blocks that are not actually a face seem to be more likely to face than correct blocks. This is due to variations in light or background reflections. In this paper, the authors propose a technique of quick comparison with a standard based on an optimized sum of absolute differences (OSAR) algorithm instead of using NEC to reduce the effects of such variational problems. During the experiments, a series of similarity measurement tests were carried out to confirm the high accuracy of using the OCAR method in comparison with other methods. To assess the accuracy of the application of the described method, two databases of persons were used - Yale Database and MIT-CBCL. The accuracy of localization was 100%.
Introduction
Localization of the face is the first step in the automatic facial recognition system. In the task of localizing a face, it is already known that a face exists in the input image, and the task is to determine the location of that face. However, locating a face from the input image is a rather difficult task due to variations in scale, posture, darkening, lighting, facial expressions, and background reflections. Despite numerous techniques for solving this problem, the task of improving localization remains extremely relevant. Research in the field of facial recognition and some techniques are presented in [1]. A recent recognition study was done by Yang et al. [2]. This group of authors divides the detection methods into 4 main categories: based on the methods of artificial intelligence [3], the invariant method for determining the characteristic points [4], comparison by the standards [5], the method of appearance [6]. Wide mapping methods are used to locate faces on input images. When using this technique, initially faces (mostly full-face images) are pre-detected and stored in a database. Later, the correlation is calculated between the blocks of the input image and the previously stored standards. The advantages of this method are its low sensitivity to noise, ease of use and it does not require a long time to locate a person from the input images [7].
However, this is not enough for the detection of faces in images with strong background and lighting variations, since these changes can change the characteristic forms of recognizable objects. The simplest method of comparison by reference is to create an average reference from a set of images stored in a database. As a result, rectangular blocks in the input image with high correlational similarity are proposed as a definition of the location of the face. This method can be called the technique of filtered comparison with the average person as a filter.
One of the widely used methods for calculating the correlation between the average face standard and rectangular blocks of input images is the similarity measurement technique, for example, the normalized cross-correlation function (NEC) [8,9]. However, the luminance indices and background reflections often affect the NIR; On images often there are blocks that are not part of the face, which have the same parameters as the standard matrix of the averaged face. This problem can be solved with the help of the sum of absolute differences (ATS) algorithm [10], which is often used to compress images and track objects, but ATS still requires additional optimization to calculate a more accurate position of the face in the image. Moreover, ATS can give a high accuracy of localization of the face in an image with high illumination, but background noise may affect the accuracy rate. The authors of the work propose an improved localization technique based on OSAR to reduce the effect of the above variations ...
[1] R. Chellappa, CL Wilson and S. Sirohey, "A survey, Human rights machine," Proceedings of the IEEE, vol. 83, pp. 705-741, 1995.
[2] Ming-Hsuan Yang, DJ Kriegman and N. Ahuja, "Detecting faces in images: a survey," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 24, pp. 34-58, 2002.
[3] Zhao Fei and Qiao Qiang, “Face detection based on the rule of law and face structure,” in Information Science and Engineering (ICISE), 2009 1st International Conference on, 2009, pp. 1235-1239.
[4] S. Jeng, HYM Liao, CC Han, MY Chern and YT Liu, "Facial feature detection using an efficient approach," Recognit, vol. 31, pp. 273-282, 3, 1998.
[5] AK Jain, Y. Zhong and M. Dubuisson-Jolly, “Deformable template models: A review,” Signal Process, vol. 71, pp. 109-129, 12/15, 1998.
[6] M. -. Yang, N. Ahuja and D. Kriegman, "Face recognition using kernel eigenfaces," in Image Processing, 2000. Proceedings. 2000 International Conference on, 2000, pp. 37-40 vol.1.
[7] Lingmin Meng and TQ Nguyen, “Frontal face localization using discriminant,” in Signals, Systems, and Computers, 1999. Conference Record on Thirty-Third Asilomar Conference on, 1999, pp. 745-749 vol.1.
[8] D. Tsai and C. Lin, “Fast normalized cross correlation for the defect detection,” Pattern Recog. Lett., Vol. 24, pp. 2625-2631, 11, 2003.
[9] Shou-Der Wei and Shang-Hong Lai, "Fast Cross-Correlation With Adaptive Multilevel Winner Update," For Image Processing, IEEE Transactions on, vol. 17, pp. 2227-2235, 2008.
[10] MJ Atallah, "Faster measure of differences," Image Processing, IEEE Transactions on, vol. 10, pp. 659-663, 2001.
Full article in English - www.ijcaonline.org/volume18/number8/pxc3872912.pdf
We are waiting for you at our webinar on February 11 at 18:00 (MSK)!
As you probably know, Innopolis University is starting a series of webinars with our teachers. The first webinar on Artificial Intelligence will be held on February 11, 2014 at 18:00 Moscow time. He will be held by Associate Professor of Innopolis University Samir Belhauari .
Registration link - attendee.gotowebinar.com/register/6601261461187578113 . Hurry up to register today, the number of places is limited!
')
Partial translation of the article on the topic of facial recognition, published by S. Belkhauari in the International Journal of Computer Applications, read under the cut.
Professor Belhauari sends you his greeting:
The article was published in International Journal of Computer Applications (0975 - 8887), Volume 18– No.8, March 2011.
The method of localization of persons using the method of quick comparison with the standard based on the algorithm optimized sum of absolute differences, N. Dawoud, S. Belhaouari, J.Janier
annotation
Until recently, the method of recognition by comparing patterns (standards) was widely used to solve the problems of localization of faces in images. The normalized cross-correlation function (NEC) is considered a measurement method traditionally used to calculate the similarity between existing face patterns and rectangular blocks of an input image to localize the position of a face in a picture. However, there is always an error in determining the position of a face due to the fact that some blocks that are not actually a face seem to be more likely to face than correct blocks. This is due to variations in light or background reflections. In this paper, the authors propose a technique of quick comparison with a standard based on an optimized sum of absolute differences (OSAR) algorithm instead of using NEC to reduce the effects of such variational problems. During the experiments, a series of similarity measurement tests were carried out to confirm the high accuracy of using the OCAR method in comparison with other methods. To assess the accuracy of the application of the described method, two databases of persons were used - Yale Database and MIT-CBCL. The accuracy of localization was 100%.
Introduction
Localization of the face is the first step in the automatic facial recognition system. In the task of localizing a face, it is already known that a face exists in the input image, and the task is to determine the location of that face. However, locating a face from the input image is a rather difficult task due to variations in scale, posture, darkening, lighting, facial expressions, and background reflections. Despite numerous techniques for solving this problem, the task of improving localization remains extremely relevant. Research in the field of facial recognition and some techniques are presented in [1]. A recent recognition study was done by Yang et al. [2]. This group of authors divides the detection methods into 4 main categories: based on the methods of artificial intelligence [3], the invariant method for determining the characteristic points [4], comparison by the standards [5], the method of appearance [6]. Wide mapping methods are used to locate faces on input images. When using this technique, initially faces (mostly full-face images) are pre-detected and stored in a database. Later, the correlation is calculated between the blocks of the input image and the previously stored standards. The advantages of this method are its low sensitivity to noise, ease of use and it does not require a long time to locate a person from the input images [7].
However, this is not enough for the detection of faces in images with strong background and lighting variations, since these changes can change the characteristic forms of recognizable objects. The simplest method of comparison by reference is to create an average reference from a set of images stored in a database. As a result, rectangular blocks in the input image with high correlational similarity are proposed as a definition of the location of the face. This method can be called the technique of filtered comparison with the average person as a filter.
One of the widely used methods for calculating the correlation between the average face standard and rectangular blocks of input images is the similarity measurement technique, for example, the normalized cross-correlation function (NEC) [8,9]. However, the luminance indices and background reflections often affect the NIR; On images often there are blocks that are not part of the face, which have the same parameters as the standard matrix of the averaged face. This problem can be solved with the help of the sum of absolute differences (ATS) algorithm [10], which is often used to compress images and track objects, but ATS still requires additional optimization to calculate a more accurate position of the face in the image. Moreover, ATS can give a high accuracy of localization of the face in an image with high illumination, but background noise may affect the accuracy rate. The authors of the work propose an improved localization technique based on OSAR to reduce the effect of the above variations ...
[1] R. Chellappa, CL Wilson and S. Sirohey, "A survey, Human rights machine," Proceedings of the IEEE, vol. 83, pp. 705-741, 1995.
[2] Ming-Hsuan Yang, DJ Kriegman and N. Ahuja, "Detecting faces in images: a survey," Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 24, pp. 34-58, 2002.
[3] Zhao Fei and Qiao Qiang, “Face detection based on the rule of law and face structure,” in Information Science and Engineering (ICISE), 2009 1st International Conference on, 2009, pp. 1235-1239.
[4] S. Jeng, HYM Liao, CC Han, MY Chern and YT Liu, "Facial feature detection using an efficient approach," Recognit, vol. 31, pp. 273-282, 3, 1998.
[5] AK Jain, Y. Zhong and M. Dubuisson-Jolly, “Deformable template models: A review,” Signal Process, vol. 71, pp. 109-129, 12/15, 1998.
[6] M. -. Yang, N. Ahuja and D. Kriegman, "Face recognition using kernel eigenfaces," in Image Processing, 2000. Proceedings. 2000 International Conference on, 2000, pp. 37-40 vol.1.
[7] Lingmin Meng and TQ Nguyen, “Frontal face localization using discriminant,” in Signals, Systems, and Computers, 1999. Conference Record on Thirty-Third Asilomar Conference on, 1999, pp. 745-749 vol.1.
[8] D. Tsai and C. Lin, “Fast normalized cross correlation for the defect detection,” Pattern Recog. Lett., Vol. 24, pp. 2625-2631, 11, 2003.
[9] Shou-Der Wei and Shang-Hong Lai, "Fast Cross-Correlation With Adaptive Multilevel Winner Update," For Image Processing, IEEE Transactions on, vol. 17, pp. 2227-2235, 2008.
[10] MJ Atallah, "Faster measure of differences," Image Processing, IEEE Transactions on, vol. 10, pp. 659-663, 2001.
Full article in English - www.ijcaonline.org/volume18/number8/pxc3872912.pdf
We are waiting for you at our webinar on February 11 at 18:00 (MSK)!
Source: https://habr.com/ru/post/212039/
All Articles