Research On Image Captcha Recognition Algorithm Based On Deep Learning

The verification code is a completely automated Public Turing Test to Tell Computers and Humans Apart(CAPTCHA).It is an important human-machine distinction technology for websites to prevent attacking from automatic malicious programs.Websites are sometimes maliciously accessed by automated programs in a short period of time,which will waste a lot of network resources.In this case,the verification code came into being.Under normal circumstances,major websites will automatically generate verification codes for users to identify.This identification process is difficult for computers,but easy for humans.If the success rate of humans in cracking the verification code can reach 90% or higher,while the success rate of the computer program is less than 1%,then this type of verification code can be considered successful.At present,all major websites have designed a variety of verification codes with low resolution,multiple noise points,deformed characters,and glued characters.In this way,a greater workload and work pressure are generated for search engine optimization.Based on this,this paper proposes an image verification code recognition algorithm based on deep learning to realize the automatic recognition of verification codes and provide convenience for search engine optimizers and testers.In the field of traditional image processing,verification code recognition technology is divided into image preprocessing,positioning,character segmentation,character recognition and other steps.Today,deep learning networks are widely used in scientific research.Although several verification code recognition algorithms use traditional digital image processing technologies in their implementation,these technologies still have shortcomings(such as weak feature extraction capabilities,susceptibility to noise in the input image,insufficient use of spatial information,etc.).Therefore,These technologies are gradually being replaced by powerful deep learning methods.The main research contents of this paper are as follows::(1)A complete data preprocessing process is proposed,including image graying,binarization and denoising.Highlight the key areas in the captcha image,reduce the impact of noise,and provide a data basis for subsequent algorithms.(2)Explore network models based on convolutional neural networks,including ResNet18 and Xception models.In ResNet,multiple loss functions are introduced for joint to improve the ability of the model.In order to more effectively extract the spatial characteristics of the verification code,the spatial attention residual module is introduced into the Xception model,and the weights of different regions are allocated to strengthen Identify the weight of key areas and reduce the influence of interference information.(3)Explored the ability of combining CNN in the image field and Transformer model in the field of sequence data in the verification code recognition.The verification code recognition is actually the process from image recognition to sequence determination.The experimental results show that this method can be effectively applied in verification code identification problem.