An Algorithm for Segmentation of CAPTCHA Characters Based on Color-Clustering and Feedback Mechanism

Abstract

Optical character recognition (OCR) is a traditional task in image processing and pattern recognition. In recent years, some new technologies of character recognition have emerged, which greatly improved the effect of OCR. As the foundation of the task of character recognition, whether the characters can be split accurately is the key point. Many researchers have done a lot of work in the car license plate segmentation and recognition, but there have been little work in the segmentation and recognition for CAPTCHA. In this paper, we propose a novel approach of segmentation for CAPTCHA characters based on color-clustering. At first, the color features of the characters are used to cluster. Then we select the effective clusters and drop the noisy clusters. The segmentation results after some post-operations are applied in the BP neuron networks. The recognition results react up on the segmentation system in order to segment the CHAPTER characters more accurately. This method can effectively deal with the CAPTCHA with complex color features and improve the segmentation accuracy by the feedback mechanism. It has strong adaptability and portability.