基于对抗样本生成的验证码反爬虫机制研究  

Study on the verification code anti-crawler mechanism based on the generation of adversarial samples

作　　者：马军 王效武[2] 朱永川 王海兮[2] MA Jun;WANG Xiaowu;ZHU Yongchuan;WANG Haixi(Shenzhen CyberAray Technology Corporation,Shenzhen 518000,China;The 30th Institute of China Electronics Technology Corporation,Chengdu 610041,China)

机构地区：[1]深圳市网联安瑞网络科技有限公司,广东深圳518000 [2]中国电子科技集团公司第三十研究所,四川成都610041

出　　处：《应用科技》2021年第6期45-50,共6页Applied Science and Technology

摘　　要：为提升网站验证码的安全性,提出基于对抗样本生成的验证码反爬虫机制。本文通过在对抗样本数据集中添加极小的扰动,可导致验证码识别模型输出错误的预测结果,从而无法绕过反爬机制对网络数据进行非法下载。针对常用的验证码识别模型,本文对比了使用图像加扰和未使用图像加扰情况下的文本验证码识别效果。结果表明,采用本文提出的图像加扰算法,可大幅度降低图像识别模型的识别精度,从而有效保护网站验证码反爬机制的可靠性。基于本文提出的图像加扰验证码技术,可作为互联网反爬虫机制的重要手段。In order to improve the security of website verification codes,a verification code anti-crawler mechanism based on adversarial sample generation is proposed.In this paper,by adding very small disturbance to the adversarial sample data set,incorrect prediction results can be output by the verification code recognition model,which refrains people from illegally download network data by bypassing the anti-crawl mechanism.Aiming at the commonly used verification code recognition model,this paper compares the recognition effect of text verification code in the cases of using image scrambling and not using image scrambling.The results show that the image scrambling algorithm proposed in this paper can greatly reduce the recognition accuracy of the image recognition model,thereby effectively protecting the reliability of the website verification code anti-climbing mechanism.Image scrambling verification codes based on this paper can be used as an important means of practicing Internet anti-crawler mechanism.

关 键 词：验证码识别 字符分割 深度神经网络 对抗样本 图像扰动 图像识别 深度学习 人工智能 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]