Abstract.

In this paper, an expansion of convolutional neural network (CNN) input features based on Hough Transform is proposed. The idea of the approach is that morphological contrasted and Hough transformed image will be used as an input for some convolutional filters. Thus, CNNs computational complexity and the number of units are not affected. Morphological contrasting and Hough Transform are the only additional computational expenses of introduced CNN input features expansion. Proposed approach was demonstrated on the example of CNN with very simple structure using two image recognition problems: object classification on CIFAR-10 and printed character recognition on private dataset with symbols taken from Russian passports. Suggested approach allowed to reach noticeable accuracy improvement without taking much computational effort, which can be extremely important in industrial recognition systems or difficult problems utilizing CNNs, like pressure ridge analysis and classification.

Keywords:

convolutional neural networks, Hough Transform, feature extraction

PP. 83-88.

REFERENCES

1. P. Hough, “Method and means for recognizing complex patterns,” (Dec 1962).
2. D. H. Ballard, “Readings in computer vision: Issues, problems, principles, and paradigms,” ch. Generalizing the Hough Transform to Detect Arbitrary Shapes, 714–725, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA (1987).
3. T. Khanipov, I. Koptelov, A. Grigoryev, E. Kuznetsova, and D. Nikolaev, “Vision-based industrial automatic vehicle classifier,” (2015).
4. N. Skoryukina, D. P. Nikolaev, A. Sheshkus, and D. Polevoy, “Real time rectangular document detection on mobile devices,” (2015).
5. D. Krokhina, V. Blinov, S. Gladilin, I. Tarhanov, and V. Postnikov, “Fast roadway detection using car cabin video camera,” (2015).
6. Q. Munib, M. Habeeb, B. Takruri, and H. A. Al-Malik, “American sign language (asl) recognition based on hough transform and neural networks.,” Expert Syst. Appl. 32(1), 24–37 (2007).
7. G. Castellano and M. B. Sandler, “Handwritten digits recognition using hough transform and neural networks,” in Circuits and Systems, 1996. ISCAS ’96., Connecting the World., 1996 IEEE International Symposium on, 3, 313–316 vol.3 (May 1996).
8. J. Koh, K. Mehrotra, C. K. Mohan, and S. Ranka, “Korean character recognition using neural networks,” tech. rep., Syracuse University, Electrical Engineering and Computer Science (05 1990).
9. C. Jennings, “Character recognition using the hough transform,” tech. rep., University of Calgary (March 1993).
10. D. Nikolaev, S. Karpenko, I. Nikolaev, and P. Nikolayev, “Hough transform: Underestimated tool in the computer vision field,” in Proceedings of the 22th European Conference on Modelling and Simulation, 238––246 (2008).
11. CIFAR-10 dataset, https://www.cs.toronto.edu/.
12. E. Kuznetsova, E. Shvets, and D. Nikolaev, “Viola-Jones based hybrid framework for realtime object detection in multispectral images,” in Proc. SPIE 9875, Eighth International Conference on Machine Vision, 98750N (December 8, 2015), (2015).
13. V. Vanhoucke, A. Senior, and M. Z. Mao, “Improving the speed of neural networks on CPUs,” in Deep Learning and Unsupervised Feature Learning Workshop, NIPS 2011, (2011).
14. R. Rigamonti, A. Sironi, V. Lepetit, and P. Fua, “Learning separable filters,” in Conference on Computer Vision and Pattern Recognition, (2013).
15. E. Limonova, D. Ilin, and D. Nikolaev, “Improving neural network performance on simd architectures,” in Eighth International Conference on Machine Vision, 98750L–98750L, International Society for Optics and Photonics (2015).
16. E. Limonova, A. Sheshkus, and D. Nikolaev, “Computational optimization of convolutional neural networks using separated filters architecture,” International Journal of Applied Engineering Research 11(11), 7491–7494 (2016).