A. Yu. Aksenov, A. A. Zaitseva, Shannaq Boumedyen
The development of computer technologies allows its application in wide variety of fields including automatic character recognition. This complex task of real-time search, localization and identification of characters in motion objects. The analysis shows that known systems firstly give unsatisfactory results on low-quality images, secondly they doesn’t work on complex images due to the positioning of recognition region problem and thirdly all of them are oriented on strictly given conditions (lightning, camera angle, brightness etc.).
The idea of proposed method of character recognition without preliminary learning consist in search of bounded pixels groups with similar values of color components based on rang distribution methods. Despite histogram distribution in rang representation the subsets of similar brightness elements are arranged not by brightness values but by frequency of its values occurrence. In simple case when all regions of brightness differ by number of elements this representation is equivalent to reordered histogram and can be referred as rang distribution. It is important to underline that not only brightness but any other image property can be chosen for rang representation, for example color. Besides, one brightness characteristic can be substituted by another one using for example equation of some initial brightness values by quantization of brightness scale (and the colors can be replaced by merging). By-element values of chosen characteristic can be defined accounting the brightness (color) values of nearby elements.
Rang distribution formation in YUV (brightness + color difference) color space is used for algorithm of text regions localization to function providing accuracy of character edges detection.
The adaptive choosing of character/background difference threshold makes this method less sensitive to lightning, exposition and color variations. Proposed method can be effectively applied for automobile license plate identification.
Complementary post processing including restrictions on possible character set and schemes of characters positioning allow excluding “false” symbols localization.