Abstract: In order to verify the safety of the current CAPTCHA on the Internet banking, seven categories of representative CAPTCHAs of the current seven banks on the Internet are chosen to break. It presents an efficient method for solving visual textual CAPTCHAs using image processing techniques and SVM, such as graying, thresholding, interference noises removing, segmentation, character normalization, extracting feature vector for each character, and recognizing it based on SVM. It discusses defense from attacking viewpoint to improve the design of visual textual CAPTCHAs.

Key words: Completely Automated Public Test to tell Computers and Humans Apart（CAPTCHA）, Support Vector Machine（SVM）, image processing

摘要： 为了评价当前网上银行登陆系统验证码的安全性，选取了其中七类代表性的验证码进行识别。对验证码进行灰度化、二值化、去除干扰、分割、归一化，进行特征向量提取，采用基于支持向量机的方法进行识别。在分析识别结果的基础上，提出了改进网上银行验证码的一些建议。

关键词: 验证码, 支持向量机（SVM）, 图像处理