Research On The Algorithm Of Palmprint Feature Extraction For Mobile Devices In Complicated Background

With the rapid development of Internet and mobile devices,also the financial payment security needs to be improved,biometrics technology has been widely used in the field of mobile payment.Compared with fingerprint,face,and voiceprint,palmprint is rich in texture information,stable in features,and easy to acquisition.Palmprint authentication technology has become a central research for mobile payment authentication.However,the complex background,variable illumination,and uncontrollable gestures at the mobile device have brought challenges to the description of the palmprint feature,the accuracy of the palmprint feature directly affects the authentication rate or the recognition rate.Therefore it is significant to study the algorithm of palmprint feature extraction for mobile devices in complicated background.In this paper,the effects of illumination,shadow,translation and distortion on the palmprint feature extraction are considered,and presents a multi-feature fusion algorithm for palmprint on mobile devices.The algorithm combines the palmprint texture features and edge features,maximizes the complementarity of texture information and edge information to accurately describe palmprint information.The multi-feature fusion algorithm is embedded into the feature extraction and matching module of the mobile palmprint authentication system,and the accuracy of the software's authentication function is tested.Aiming at the characteristics of low computing resources and slow computing speed on mobile devices,proposes a method of palmprint feature extraction based on convolutional neural network and an architecture of mobile and server working together.The mobile device connects the network,and then requests the training model on the server to extract palmprint feature and reduce the amount of computation.Using VGGNet-16 as the initial model to adjust the iterative speed of momentum,reduces the value of error function to the target range in a short time,so as to shorten the time of training model.Customizing dimension of the fully connected layer fc-6 to ease overfitting,also improves the recognition accuracy and training speed.Establishing a network structure based on Triplet Loss function,this network structure directly learns the mapping from image to Euclidean embedded space,it enables the L2 distance in directly correspond to the similarity of palmprint,describes the palmprint feature more efficiently and accurately and improves the recognition accuracy.In this paper,proposes a multi-feature fusion algorithm based on transform domain and a feature extraction algorithm based on convolutional neural network,which eliminate the miscellaneous information caused by such factors as illumination,rotation,translation and blurring,achieving information compression and reducing computation.Through simulation experiments,the effectiveness of the above algorithms are verified.