Research On Captcha Recognition Technology With Hidden Markov Model

CAPTCHAs with poor security and weak resistance to attack are used by many websitesand applications currently, but the systematic and standard design and assessment forCAPTCHAs do not yet exist. The CAPTCHA image recognition technology is related toimage processing, pattern recognition and many other fields. With more in-depth research wecan not only improve the security of the websites, but also help to promote the artificialintelligence technology.We focus on recognition technology of text-based CAPTCHAs and make a systematicresearch with Hidden Markov Model. First we mainly make a study of the CAPTCHAs withmore serious distortion and interference in detailed analysis, which are difficult to berecognized. Then we introduce a suitable pseudo2D Hidden Markov Model based onstructural features of the CAPTCHAs, and combine with a series of image recognitionprocesses, which include image denoising and segmentation involved in the preprocessingsteps, to build an integral system based on Hidden Markov Model. The system is expected tobe more efficient and accurate for CAPTCHA recognition. Finally, through simulation wecompare it with other common recognition technologies, and the results show that the pseudo2D model is more accurate indeed than the traditional model. In the next we propose a pseudo2D model with rough classification and prove that it is not only well adaptive and effectivebut also faster to finish the recognition task.The Hidden Markov Model has good characteristics to describe both local features andoverall structures of objects. There is no specific research on CAPTCHA recognition usingHidden Markov Model at present, but we successfully apply it to recognize complexCAPTCHAs at last and obtain well recognition performance. So our recognition method withpseudo2D Hidden Markov Model in the field is innovative to some extent and also providesvaluable reference for safer and more reliable CAPTCHA design to some degree.