Research On The Key Techonologies Of The Captcha With Adhersion Characters Identification

The early 21st century,the Internet has developed greatly,a variety of web services that could make people's lives richer and easier emerged.The network security issues have become more and more concerned by the service companies and people.As one of the most common basic means to guarantee network security,the CAPTCHA technology has been widely used and developed by many scientific researchers.In one of the research direction of it,the CAPTCHA identification technology analyzes and discovers existing defects of CAPTCHA system,improve the security.CAPTCHA identification technology is a comprehensive study and application of the pattern recognition theory and digital image processing theory.This paper introduces the technology and theories of every processing stage in CAPTCHA identification,including image preprocessing stage,the image characters segmentation and the character recognition stage.This paper has also designed and implemented the complete recognition methods and achieved good performance.First of all,I analyze and study the technology of digital image processing used in image preprocessing in depth,focusing on the theory and common technical means of important stages,including the image grayscale,binary processing,image noise cancellation,disturbance removal and so on.This paper has designed an appropriate solution for the curved lines in CAPTCHA,which obtained good results.Secondly,I introduce a variety of methods and theories of the image characters segmentation,and design the implement steps for each algorithm,analyze the advantages and disadvantages of various methods and applicable scenario.The improved connected component detection algorithm is proposed in this paper,which can adapt the mild adhesion of character segmentation well.On the basis of the connected area detection and projection method,we designed a segmentation approach based on connected area detection and projection by adding excessive segmentation post-processing strategy,for the presence of pixel loss after preprocessing scene has a good adaptability.Finally,I analyze and study three kinds of character modeling method and theory,and illustrate their advantages and adaptable scenes.For the feature modeling,this paper proposes a feature based on contour edge trend information,which has a good adaptability of the rotation and deformation characters.This article implements all the process of the recognition in Java language.Through the experimental analysis,the paper proves that the strategy and the method designed in this paper achieve good recognition results.At the same time,the feature model in this paper has a good character description capability.