Research Of Non-contact Palmprint Recognition Technology Based On Deep Convolutional Network

With the development of information technology and the popularization of mobile devices,contactless palmprint recognition technology is considered to be an effective method for mobile device authentication.Compared with traditional palmprint recognition application scenarios,mobile palmprint recognition is susceptible to complex backgrounds,illumination,palm gestures and other factors.The research on non-contact palmprint recognition technology for mobile terminals is a hot issue in current research.In recent years,the rapid development of deep learning technology has provided effective technical means to mobile palmprint recognition.This paper uses deep learning technology to study the key issues in mobile palmprint recognition.The main work is as follows:(1)Propose a mobile terminal palmprint anti-spoofing detection method based on Faster RCNN and establish a palmprint anti-spoofing detection database.In view of the possible image and video attack problems in mobile palmprint recognition,propose the use of Faster RCNN network for palmprint anti-spoofing algorithm research.It is difficult to distinguish the real palmprint image from the fake image in the RGB color space.In this paper,the YCb Cr space image with chroma and brightness information is used for network training,and the color information and texture features are fused to confirm the identity and ROI locating.And make detection and locating more accurate by adjusting the anchor scale in the network.Achieve better detection results on the in-house database for palmprint anti-spoofing detection problems.The method proposed in this paper can resist image attacks with an efficiency of 99.5%.(2)Aiming at the problem of limited hardware resources of mobile terminals,a lightweight palmprint recognition convolutional neural network PR-Res Net is proposed.Based on the Residual Network,an inverted residual block is used to construct a deep network for feature extraction to reduce feature degradation,while using deep separable convolutions instead of standard convolution in the neural network.Perform model optimization to reduce parameter storage and computational complexity.Experimental results show that the method performs well on two public databases(CASIA,IITD)and one in-house database for mobile terminal palmprint recognition.The method proposed in this paper occupies less storage resources and the recognition accuracy rate reaches 95.77%.(3)Design and implement a palmprint authentication system on mobile devices based on TFlite.The palmprint authentication system for mobile devices is mainly divided into three steps.The palmprint image collection is completed under the guidance of the palm guide line by calling the camera.Click the registration button to complete user registration use the palmprint recognition algorithm.And click the authentication button to complete the identity authentication through the feature matching algorithm.In this paper,we have research on non-contact palmprint recognition technology based on deep convolutional network.Design a lightweight PR-Res Net network,and propose a mobile terminal palmprint anti-spoofing detection method based on Faster RCNN use color information and texture features.Design and implement a palmprint authentication system on mobile devices based on TFlite.