Study On The Security Or Non-adhesive Character CAPTCHA

CAPTCHA,as a method to distinguish between a human user and a computer user,has played an important role for network securities.We may find drawbacks in the design of CAPTCHAs through the study of the text-based CAPTCHA recognition technology,and make suggestions for improvement of the design scheme,so as to improve the securities of CAPTCHAs and the networks.In general,we consider that a CAPTCHA is safe if the recognition rate for humans is greater than 90%and the recognition rate for computers is less than 10%.In this thesis,based on digital image-processing and pattern recognition methods,a computer identification method for text-based CAPTCHAs was developed.The main work and achievements of this paper is as follows:(1)For six kinds of CAPTCHAs with regular characters,a character segmentation algorithm of connected components was proposed based on an edge detection method.A template dataset with 1384 images were established,including numbers,letters and some Chinese characters.A template matching method was used for character recognition.The recognition accuracy was more than 90%.(2)For CAPTCHAs with non-regular characters,a BP neural network was used for character recognition,which was trained with the Chars74K data set and samples of characters segmented from the CAPTCHAs.The recognition accuracy was more than 90%.(3)To analyze the key technology in CAPTCHA recognition,a design scheme of CAPTCHA recognition was proposed.Cross-platform software for CAPTCHA recognition was designed and developed,which had main functions including image preprocessing,denoising,edge detection,character segmentation and character recognition.Using an improved template matching method and BP neural network to recognize the seven kinds of CAPTCHA images we achieve an improved recognition rate.Security vulnerabilities of the seven kinds of CAPTCHAs were identified.Suggestions for improving security of CAPTCHA were put forward in the end.