验证码机制的鲁棒性和可用性研究

项目名称： 验证码机制的鲁棒性和可用性研究

项目编号： No.61472311

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 高海昌

作者单位： 西安电子科技大学

项目金额： 82万元

中文摘要： 验证码，作为一种能自动区分计算机程序和人类用户的公开图灵测试，广泛应用于各大商业网站。但实际应用中的绝大多数验证码都存在严重的鲁棒性(如很多已被成功破解)和可用性(如验证时间长，识别成功率低)问题。本课题拟在前期成功破解空心和粘连验证码、提出两种新型图像和语音验证码的基础上，继续在以下几个方向开展工作：(1)提出通用的文本验证码自动识别策略，要能够成功识别Alexa排名网站前20的主流验证码；研究不拆分整体识别的新思路和基于字根的中文验证码识别算法，并提出改进建议；(2)对已有的图像和语音验证码进行自动识别，并提出新型的图像和语音验证码替代机制；(3)设计可以抵御MITM和众包攻击的新型协议，能够有效区分普通用户和AMT肉机的策略。本课题研究内容属于图像处理、模式识别和机器学习等多个学科交叉领域，具有重要的理论和应用价值。希望通过本课题的研究，全面提升验证码的鲁棒性和可用性。

中文关键词： 网络服务安全；访问控制；验证码；人机交互验证；机器学习

英文摘要： As a completely automated public Turing test to tell computers and humans apart, CAPTCHA has found widespread applications on numerous commercial web sites. But the most widely used CAPTCHAs have serious problems both in robustness (many CAPTCHAs have been broken successfully) and usability (long verification time and low success recognition rate). On our previous works, we have broken the hollow and CCT CAPTCHAs successfully, and proposed two new image and audio CAPTCHAs. We are going to do further research in the following directions: (1) Propose general text-based CAPTCHAs cracking strategy which can recognize all the top 20 website CATPCHAs in Alexa. Study new idea based on no-decomposition and new Chinese CAPTCHA recognition algorithm based on Chinese etymon, and offer improvement advices; (2) Identify existing image and audio CAPTCHAs automatically and design new image-based and audio-based CAPTCHA mechanisms; (3) Design new protocol to defend against MITM and crowdsourcing attacks, and propose effective countermeasure to distinguish general users from Amazon Mechanical Turks. The research contents of our subject, involving many interdisciplines such as image processing, pattern recognition and machine learning, have important significance of both theoretical and practical values. We hope to improve the robustness and usability of CAPTCHAs greatly in the general through our study.

英文关键词： Web services security;Access control;CAPTCHA;Human interactive proof;Machine learning