Research And Implementation Of Identity Recognition Algorithm Based On Signature Hand Features

The rapid development of the Economy and Internet industry has brought the payment industry booming,among them,the credit card to pay because of its convenience,fast features such as payment has become a consumer daily life consumption,a major means of financial transactions.It is extremely important to ensure the security of credit card transactions in the case of large amount of credit card transactions and transactions.Card swiping fraud is the most important security threat facing card swiping at present.Therefore,rapid and accurate identification of card traders is an effective means to prevent fraud detection.At present,our country swipes the card transaction to take the password way as the authentication swipes the card user's identity main means,but,relies on the password only cannot completely effectively prevent swipes the card fraud.Therefore,in order to guarantee the safety of the credit card payment,no matter how password-less authentication,all credit card transactions require users to sign,signature and the identity of the user specific characteristics,such as hand features to the user's identity authentication recognition is the most natural,is the most widely accepted solution.In this paper,video algorithm of user signature action in natural scene is designed to identify the user's identity according to the user's specific human characteristics.This paper mainly studies the following three aspects.(1)Construction signature video data set and detection of hand region.In this paper,the signature video data set which meets the requirements of research is first constructed to provide data support for the subsequent research.At the same time,in the signed video,under different environment,illumination and other conditions,the captured video frame has many factors,such as too much background area and too much background noise,which lead to poor detection effect of hand region.Therefore,the detection algorithm is used to extract the hand region with a small amount of background.In this paper,the detection method based on HOG feature and SVM classifier is applied respectively,the detection algorithm based on DPM and the detection algorithm based on depth learning are used to extract hand region.(2)Segmentation and extraction of hand region.In this paper,extraction of hand region is regarded as the semantic segmentation task of two categories.In this paper,a full convolution neural network(CNN)model based on the attention mechanism and dilated convolution module is designed,which mimics the attention mechanism of human vision and use DCB module to extract multi-scale feature.So that the model can pay more attention to the significant region in the convolution feature.The experimental results on three common data sets and the data sets of this paper show that the proposed algorithm can achieve a good result of hand region segmentation.(3)Feature extraction and identity recognition.In order to extract the local feature of the hand region and the global feature of the whole video frame,the local feature of the hand region and the global feature of the whole video frame can be extracted and learned from the extracted local feature of the hand region and the global feature of the video frame.In this paper,two twinning CNN models with two branches are proposed: early fusion based identity recognition algorithm and late fusion based identity recognition algorithm.Different fusion methods are used for hand local feature fusion and static video frame feature fusion,respectively.Finally,the user identification is realized.