CAPTCHA Recognition Based on Two Dimensional RNN
Ru Chen · Journal of Chinese Computer Systems · 2014
Completely automated public turning test to tell computers and humans apart( CAPTCHA) is a kind of network security technology w hich is designed based on unsolved artificial intelligence problems,CAPTCHA recognition is an important branch of this field. Recurrent neural netw ork( RNN) of Long short term memory( LSTM) has been successfully applied to CAPTCHA recognition,essentially RNN of LSTM is one dimensional,w hile Text-based CAPTCHA is tw o dimensional. This paper applied the tw o dimensional RNN of LSTM to the CAPTCHA recognition,this method combined tw o phases of feature extraction and recognition,and the context can be learned w ell by the neural netw ork. A Rejection strategy based on support vector machine is proposed to improve the reliability of the result. Experimental results show follow ing tw o points: Firstly,tw o dimensional RNN can get higher recognition rate than one dimensional RNN. Secondly,the new rejection algorithm is superior to other methods.