【摘要】 随着深度学习技术与图像处理技术的进步,验证码的有效性受到了挑战。文章分析了现有的针对滑块式拼图验证码的攻击方案,对于利用Canny边缘检测算法寻找高于阈值的特征点匹配度来定位目标位置的攻破滑块式拼图验证码的攻击方法,针对性地提出了一个抹除目标位置部分边缘特征点以降低目标位置特征点匹配分值的验证码生成算法;同时,为抵抗穷举攻击,该算法取消了滑动条,直接以拼图块作为滑块,将拼图块的移动范围从一维扩大至二维。经过安全性分析、易用性分析、人类测试以及攻击实验,该方案所生成的验证码在保证了易用性的情况下,对于特征点匹配攻击行为具有抵抗性。更多还原

【Abstract】 With the evolution of deep learning techniques and image processing techniques, the effectiveness of CAPTCHA is under the threat of being compromised. In this paper, the current researches regarding breaking and bypassing slider puzzle CAPTCHAs are discussed, and a technique to break slider puzzle CAPTCHA exploiting Canny edge detection algorithm to pinpoint target areas is analyzed. An algorithm of CAPTCHA generation involving partially removing edge features is proposed accordingly. In this algorithm, edge features of the target area are partially removed after being extracted with Canny algorithm, resulting in the reduction in feature-matching degree. The puzzle picture is considered as the slider, instead of using a horizontal slider to control it, expanding the scope of the puzzle picture from one-dimensional area to two-dimensional area, in attempt to gain resilience under brute-force attacks. With analysis of security and ease of use, human tests and attack experiments, it is proved that this algorithm is resistant to feature-matching attacks without compromise of ease of use.更多还原