基于LSTM型RNN的CAPTCHA识别方法  

CAPTCHA Recognition Method Based on RNN of LSTM

作　　者：张亮[1] 黄曙光[1] 石昭祥[1] 胡荣贵[1] 

机构地区：[1]解放军电子工程学院网络系,合肥230037

出　　处：《模式识别与人工智能》2011年第1期40-47,共8页Pattern Recognition and Artificial Intelligence

摘　　要：全自动区分计算机和人的图灵测试(CAPTCHA)是一种基于人工智能难题的网络安全机制.研究CAPTCHA的识别能够使其变得更加安全,并能促进一些人工智能难题的求解.文中首先对现有的CAPTCHA识别方法进行总结和分析,然后提出一种基于长短时记忆(LSTM)型递归神经网络(RNN)进行识别的方法,并对CAPTCHA识别中的特征提取问题进行研究.最后,为进一步提高RNN的识别率,提出一种解码算法.实验结果表明,文中方法是有效的,灰度值对于RNN是一种较好的特征,提出的解码算法能够取得较高的识别率,又有较低的时间复杂度.Completely automated public turing test to tell computers and humans apart （CAPTCHA） is a kind of network security mechanism based on hard artificial problems. Study of recognition of CAPTCHA impels it to become more secure, and some hard atifical problems to be solved. Firstly, CAPTCHA recognition methods of state of the art are analyzed. Then, a recognition method is brought up based on recurrent neural network （RNN） which is composed by long short-term memory （LSTM） blocks. Thirdly, feature extraction for CAI^CHA recognition is studied. Finally, a decoding algorithm is proposed to improve the recognition rate. Experimental results show that the proposed recognition method is efficient. Gray value of images is proved to be a kind of good feature for RNN. Furthermore, the proposed decoding algorithm gets high recognition rates with low time complexity.

关 键 词：人工智能 脱机文字识别 全自动的区分计算机和人的图灵测试(CAPTCHA) 长短时记忆(LSTM) 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]