Authentication System Based On Human Hand Characteristics In Signature Video

The financial card have become a popular payment method.However,the transaction security of bank cards should be improved.The cases of fraudulent use and forgery of bank cards are endless,even after the magnetic stripe cards which are easy to be read and written are replaced by the cards with chip.The user authentication of bank cards can only be verified by password,a six-digit array,which is too simple to guarantee the security.However,whether or not the users need to enter the password in transaction,they have to sign for confirmation.The personal signature contains a lot of characteristics of the user.And considering that the signature,as an already existing step when using the cards,doesn't add additional work for users,this paper will design an algorithm based on the signature video recorded when the card users signing to perform identity verification.The main contents of this work are as follows:(1)Extraction of signing action video frames.In the obtained signature video,there are both the video frame that the user is signing and the frame that not.Therefore,before the user identity verification,the non-signing vedio frame with irrelevant information,which may affect the final authentication result,needs to be removed.This paper proposes a key frame extraction algorithm based on the two-stream convolutional networks in deep learning to extract the video frame of the user's signing action.(2)Segmentation of hand region in signing action video frame.In the natural environment,the signing video contains not only the user's hand information,but also many interference factors.Therefore,it is necessary to segment the hand region and remove the irrelevant information in the signing action video frame.For this problem,an fast segmentation neural network based on deep learning is designed,which can remove the area that not belonged to the user's hand to ensure the validity of the identification.(3)Feature extraction and identification.After obtaining the signing action video frame and the user's hand segmentation image in writing state,in order to carry out effective and accurate identification work,it is necessary to perform good feature extraction on the image and the motion information of the human hand.In this paper,the two-stream network of the attention mechanism fuses the fine image information and the hand's motion,and finally realizes the identification of the hand video for the user's signature process.