Research On Adhesive CAPTCHA Based On KNN Technology

With the advancement of technology advances,the network gave rise to abundant resources,its security issues is very important.A validation code,effectively solve this problem,and network protection and information security,has a wide range of applications.But in data mining,the existence of the verification code and hindered the we get the data.Thus the research purpose of this paper is to identify CAPTCHA images appeared in the process of data acquisition.But with rising network attacks,and verification code technology is the continuous improvement.From the original simple,regular numbers,letters,and now a variety of graphic text,interactive,touch-type and improved numbers,characters,and other diversified verification code type.Due to the diversity of the type of authentication,one can not design a model of a universal identification codes,and only for a certain type of algorithm design codes.This paper is selected for improved digital,character type codes.The main features of this type codes are:(1)a wide variety of background noise;(2)rotating twisted character;(3)the character adhesions.School network codes used in this paper is the type most typical one.Its main contents are summarized as follows:(1)Analysis page,a verification code to collect data sets through the school network analysis CAPTCHA image,this paper proposes three primary colors RGB noising method to remove background noise verification code.At the same time,it was compared with the traditional method of de-noising background,showing good results.(2)a method for local interference noise,using adjacent pixel noise judgment will be removed.While the CAPTCHA image binarization.(3)for the character distortion,adhesion characteristics,in terms of character segmentation introduced mean cutting algorithm combines with water droplets,water droplets and the upper and lower contour difference projection algorithm combining segmentation and direct average.Recognition results and the three divided compared.(4)Finally,to give a single character model,using machine learning algorithms KNN,character recognition,resulting in the recognition results.Thesis innovation lies in:(1)first proposed the RGB color background denoising method which can effectively remove most of the background noise in the image,compared to the traditional de-noising method,a very good performance.(2)up and down with water droplets contour projection algorithm combined method,more accurate cutting out a single character.