A new segmentation method for connected characters in CAPTCHA
Pengpeng Lu, Liang Shan, Jun Li, Xunwei Liu · 2015
In the Completely Automated Public Turing test to tell Computer and Humans Apart recognition systems, character segmentation serves as a connecting link between the preceding and the following. By studying a variety of character segmentation algorithms, the improved method combined vertical projection algorithm, improved drop-falling algorithm and BP neural network classifier is proposed for merged characters. Firstly, this paper judges merged characters by the aspect ratio of connected component extracting from the images. Secondly, the division points are sought by the vertical projection minimums of connected components, and then these points are used as starting point of the improved algorithm to segment connected characters. Finally, BP neural network classifier is applied to select the best dividing line combinations. Experimental results show that this method can effectively solve the problem of merged characters segmentation.