An improved maze algorithm in contour extraction of CAPTCHAs

Abstract

An improved maze algorithm has been first advanced, combining with the rotation invariant theory, to extract the contour of CAPTCHAs which are slant and have random positions. First, take the CAPTCHAs as a maze, in which white pixels represents a potential route; Second, search towards “left-down” in the predefined direction and record the pixels which have been checked; Finally, get the clockwise rotate projection of CAPTCHAs within the paths' closed region in order to obtain the CAPTCHAs' contour features which are immune to rotation, random position, slant and other factors. Experiments have shown that the algorithm has better performance in dealing with CAPTCHAs which are slant and have random positions, its sufficient extraction of contour features reduces errors and improves recognition rate.