ISSN 2079-3537      

 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
Scientific Visualization
Issue Year: 2016
Quarter: 3
Volume: 8
Number: 3
Pages: 65 - 77
Article Name: APPLICATION OF THE VIOLA-JONES ALGORITHM, HOUGH TRANSFORM AND ARTIFICIAL NEURAL NETWORKS FOR RECOGNITION CHARACTER INFORMATION ON A COMPLEX BACKGROUND
Authors: A.A. Druki (Russian Federation), P.À. Kakovkin (Russian Federation), D.S. Cherneta (Russian Federation)
Address: A.A. Druki
druki2008@yandex.ru
National Research Tomsk Polytechnic University, Russian Federation

P.À. Kakovkin
Dimano1993@rambler.ru
National Research Tomsk Polytechnic University, Russian Federation

D.S. Cherneta
nitrokot@mail.ru
National Research Tomsk Polytechnic University, Russian Federation
Abstract: The work is dedicated to development of algorithms to increase the efficiency of character recognition on a complex background with noise, affine and projective distortions. The paper proposes an approach to solving the problem of character recognition, which includes three main stages: detection of region character locations in the image; image normalization and selection of individual characters; character recognition. The Viola-Jones algorithm realization is proposed for detecting of the characters’ region in images with complex background. Image normalization includes the following operations: selection borders, Hough transform, closed contours search, image alignment. The search of characters string horizontal edges in the image is performed by using the Hough transform. The detection of individual characters’ regions is performed by using the closed contours search. The configuration of convolution neural network is proposed for character recognition. The testing results of the developed algorithms efficiency and their comparison with existing analogues are presented. On the basis of the approach, the software system which shows the high efficiency of character recognition on a complex background intended to recognize the car number plates in the images was developed.
Language: Russian


Open Article   Download ZIP archive