Research On CAPTCHA Recognition Technology Based On Convolutional Neural Network

The rapid development of Internet technology has brought convenience to people's lives,and the security issues involved have become increasingly prominent.The malicious behaviors such as the abuse of network resources and the theft of network information have attracted great attention from all walks of life,which makes the CAPTCHA mechanism arises at the historic moment,and is applied to each big web site to prevent the malicious attacks of the computer program,maintain the security of the network.The research of CAPTCHA recognition can find its design loopholes in time,improve the CAPTCHA design mechanism,and better guarantee network security.Most of the existing CAPTCHA recognition methods adopt the idea of first segmentation and recognition.The accuracy of such method recognition mainly depends on the effect of character segmentation.Due to the diversity of CAPTCHA types,the traditional segmentation algorithm is very complicated for some complex CAPTCHAs.It is difficult to segment,and the universality of the algorithm is poor.Therefore,there is no universal CAPTCHA recognition method.In this paper,we propose improved schemes from the points of segmentation algorithm and CAPTCHA recognition process.The specific work is as follows:1.A CAPTCHA recognition algorithm is proposed based on the fusion of convolutional neural network and neighborhood projection segmentation.Firstly,the threshold parameters are introduced to make the segmented region fall into the minimum expected neighborhood through threshold operation.The projection method is then used to select the minimum value points aiming to improve the adaptability of the segmentation algorithm for different types of CAPTCHAs.Secondly,the convolutional neural network is used to train and recognize the segmented characters,so that they have better generalization ability.The experimental results show that the algorithm has a good recognition effect for different types of CAPTCHAs.2.Designing a recognizable end-to-end convolutional neural network CAPTCHA model without character segmentation.The flow of the traditional CAPTCHA recognition algorithm is improved.The CAPTCHA images are inputted into the convolutional neural network model.Then using the learning ability of the network to extract the character feature information automatically.Training it to generate the optimal model weight and output the prediction result.The experimental results show that the end-to-end CAPTCHA recognition algorithm based on deep learning has the better versatility.And the recognition rates of the CAPTCHA with multiple complex types can reach between 96%and 99%.