Recognition And Research Of Hollow Capcha Based On Convolution Neural Network

The Internet has spread all aspects of people's lives,people are enjoying the convenience of the Internet to bring us all the time.Website management of a large number of user informations,it's particularly important to ensuring site security and effective user informations.In order to prevent the malicious use of robots or virus software automatic batch registration,malicious dissemination of spam,etc.CAPCHA(Completely Automated Public Turing Test to Tell Computers and Humans Apart)technology is used in a large number of Web site registration verification link.The research on the capcha can find out the shortcomings of the verification code design,and put forward some improvement direction to ensure the security of the website.The identification of the common solid verification code has been studied by the support vector machine(SVM),self organizing map(SOM)and neural network(NN).However,the existing character recognition methods are still needed to be improved in the recognition of hollow capcha with the simple outline information.Therefore,the identification of hollow capcha,security remains to be explored.In this paper,the algorithm of BP neural network is used to carry out experiments,through the image preprocessing,feature extraction,training network to get the recognition rate of BP neural network algorithm,making a comparison of HU matrix and full pixel matrix feature extraction method to find the optimal network to improve the recognition rate.Then,we use the algorithm of convolutional neural network in depth learning to identify the hollow capcha.This paper analyzes the influence of the depth of the convolution network,the number of convolution layers,the convolution kernel size and the number of convolution layers on the recognition results.Finally,an optimal convolutional neural network structure is selected,and a simple interface capcha recognition simulation system is designed.The experimental results show that the convolutional neural network has a high recognition rate for the identification of hollow captha.In the case of the optimal network structure,the convolution neural network algorithm in depth learning is more stable and has higher recognition rate and better performance than BP neural network algorithm in shallow network.