Bio-inspired unified model of visual segmentation system for CAPTCHA character recognition

In this paper, we present a bio-inspired unified model to improve the recognition accuracy of character recognition problems for CAPTCHA (completely automated public turing test to tell computers and humans apart). Our study focused on segmenting different CAPTCHA characters to show the importance of visual preprocessing in recognition. Traditional character recognition systems show a low recognition rate for CAPTCHA characters due to their noisy backgrounds and distorted characters. We imitated the human visual attention system to let a recognition system know where to focus on despite the noise. The preprocessed characters were then recognized by an OCR system. For the CAPTHA characters we tested, the overall recognition rate increased from 16.63% to 70.74% after preprocessing. From our experimental results, we found out the importance of preprocessing for character recognition. Also, by imitating the human visual system, a more unified model can be built. The model presented is an instance for a certain type of visual recognition problem and can be generalized to cope with broader domains.