350 rub
Journal Neurocomputers №10 for 2013 г.
Article in number:
Face detection algorithm based on the convolutional neural network
Authors:
I.A. Kalinowski - Post-graduate Student, Tomsk Polytechnic University. E-mail: kua_21@mail.ru
V.G. Spitsyn - Dr. Sc. (Eng.), Professor, Tomsk Polytechnic University
Abstract:
The problem of highlighting faces in the images or in video stream is classical in the area of image processing. Research in this area has been being carried by many scientists, but has not been yet proposed a universal algorithm to reliably detect a face for any distribution of light, various turns, tilt and scale of the face. The purpose this work is to try to make the next step towards the creation of such an algorithm. To solve this problem, a type of artificial neural networks of direct distribution was used - the convolutional neural network (CNN). There have been studies of different configurations of CNN and different algorithms of its training on a large set of training data (233.000 images). As a result, a network is built that provides a good quality of face detection. The network was compared with the one of the most common methods of searching for faces - the algorithm of Viola-Jones. Test results show that a 20% reduction of the number of positive detection, the number of false detection decreased by 76%. More faces detected by Viola-Jones algorithm, is because that it is more resistant to turns and tilts head. Significant drawback of the developed algorithm is the speed, which in 1167 times less than the speed of Viola-Jones algorithm when scanning a frame sized of 640×480 pixels.
Pages: 48-53
References

  1. Makarenko A.A. Algoritmy' i programmnaya sistema klassifikaczii polutonovy'x izobrazhenij na osnove nejronny'x setej // Dis. kand. tex. nauk: 05.13.18. Tomsk. 2007.
  2. Xajkin S. Nejronny'e seti: polny'j kurs. M.: Vil'yams. 2006.
  3. LeCun Y., Boser B., Denker J., et al. Backpropagation applied to handwritten zip code recognition // Neural Computation. 1989. № 4. P. 541-551.
  4. LFW Face Database // www.cs.umass.edu: sajt univ-ta Amherst. Massachusets, SShA. http://vis-www.cs. umass.edu/lfw/
  5. Libor Spacek-s Facial Images Databases // www.cmp.felk.cvut.cz: sajt Centra mashinnogo vospriyatiya Cheshskogo texnicheskogo univ-ta, Praga, Chexiya. http://cmp.felk.cvut.cz/~spacelib/faces/
  6. Tan X., Triggs B. Enhanced local texture sets for face recognition under difficult lighting conditions // IEEE Transactions on Image Processing. 2010. V. 19. № 6. P. 1635-1650.
  7. Vrahatis M.N., Magoulas G.D., Plaglanakos V.P. Globally convergent modification of the quickprop method // Neural Processing Letters. Netherlands: Kluwer Academic Publishers. 2000. P. 159-169.
  8. Xoang K.T. Razrabotka i issledovanie nejrosetevy'x algoritmov upravleniya // Dis. kand. tex. nauk: 05.13.01. M. 2006.
  9. Viola P., Jones M.J. Robust real-time face detection // International Journal of Computer Vision. 2004. V. 52. № 2. P. 137-154.
  10. Bradski G., Kaehler A. Learning OpenCV. Sebastopol: O-Reilly Media. 2008. P. 556.
  11. Korshunova S.G., Stepanova O.B., Tetik L.V. Sfericheskaya model' prostranstva e'moczional'nogo vy'razheniya licza, osnovannaya na vosprinimaemy'x razlichiyax //  Nejrokomp'yutery': razrabotka, primenenie. 2012. № 2. S. 42-51.
  12. Vartanov A.V., Malyasova E.A. Zritel'noe vospriyatie i kategorizacziya ob''ektov s variativnoj formoj // Nejrokomp'yutery': razrabotka, primenenie. 2012. № 5.
  13. Andreeva O.N., Brunov A.V., Oczenka zritel'noj vospriimchivosti izobrazhenij operatorom ARM v sistemax real'nogo vremeni // Nejrokomp'yutery': razrabotka, primenenie. 2013. № 8. S. 50-53.