Research On Preprocessing Algorithm Of Palmprint Recognition With Low Constraint Conditions In Mobile Environment

With the rapid development of mobile terminal devices and mobile Internet,more and more network activities are moving to mobile environment;however,the security of mobile Internet suffers from more severe technical challenges.For example,wireless communication information is easier to be intercepted,tampered,etc.Thus the security requirements in mobile environment is much higher.Accurate personal identification/authentication is the significant core and cornerstone of cyberspace security,definitely including mobile environment.However,traditional identity authentication methods for mobile environments,such as digital password and nine-point pattern,cannot completely meet the security requirements currently.As a novel identity authentication method for mobile environment with higher uniqueness,security and stability,biometrics has been gradually studied and widely used in the field of mobile security.Biometrics recognition uses individual physiological or behavioral characteristics to authenticate personal identify.Physiological modalities include fingerprint,face,iris,palmprint,etc.;behavioral modalities include signature,gait,and voice.Compared with other biometric modalities,palmprint has several advantages,including rich discriminative information,difficult leak,non-contact acquisition,etc.The cameras on mobile devices are developing rapidly,and palmprint images can be directly acquired with the built-in camera without additional hardware cost and configuration;therefore,palmprint recognition in mobile environment has gradually become hot in recent years.However,the palmprint recognition in mobile environment faces many challenges,such as complex background,illumination variance,uncontrollable hand posture and position,limited hardware resources for computation and storage.Palmprint preprocessing is a key part of palmprint recognition,and its main objective is to localize and crop the region of interest(ROI)of palm.The preprocessing accuracy directly affects the subsequent feature extraction and recognition accuracy.Thus,based on the deep learning framework,a preprocessing algorithm for palmprint recognition with low constraints in mobile environment is proposed in this dissertation.The main work and research contributions are summarized as follows:(1)Background,research significance and related knowledgeFirst,we introduced the project background and research significance,and also analyzed the feasibility of palmprint recognition in mobile environment.Secondly,we described the research status of traditional and deep-learning-based preprocessing algorithms and feature extraction algorithms for palmprint recognition,secure palmprint recognition and the development of object detection and semantic segmentation,and then pointed out the shortcomings of the existing methods and the problems to be solved.Finally,we introduced the secure palmprint recognition and the evaluation indices of recognition accuracy.(2)Low-constraint inter-finger valley(IFV)automatic detection and segmentationThe proposed preprocessing algorithm includes three parts,namely IFV detection network,triple-advantage selection criterion and IFV segmentation network.The lowconstraint conditions during hand image acquisition improve the flexibility and comfortability.A user only needs to stretch his/her hands,naturally opens five fingers,and ensures that the direction of the middle fingertip is approximately vertically upward.Under the low-constraint conditions,the IFV detection network based on deep learning was used to automatically detect the IFVs in the hand images without any auxiliary technique.The selection criterion was designed to achieve triple advantages,which could select two required IFVs,remove the false IFVs,and judge the unqualified image samples.In addition,an IFV semantic segmentation network was designed to accurately segment the hand region in IFV with strong resistance to skin-like background.In order to train the IFV detection and segmentation networks,a palmprint database and an IFV segmentation database were established and labeled.(3)Two-fold segmentation and key point detectionFailure criterion of semantic segmentation was designed.When semantic segmentation failed,the twice adaptive skin-color segmentation was launched as a twofold repair to compensate for the failure of the IFV segmentation network,and to further reduce the error rate of key point detection.The convex hull method was used to extract the hand contour in IFV,avoiding the adverse effects of unsmooth contours and singular points on key point detection.The angle bisector was obtained between two straight lines that are fitted from the contours of two fingers.The intersection between the angle bisector and the hand contour was considered as the key point,which increased the localization accuracy of the palm ROI.To sum up,to solve the severe challenges that are faced by palmprint processing in mobile environment,the proposed novel palmprint processing algorithm can significantly improve the comfortability,and substantially improve the resistance and recognition accuracy with low constraints in complex backgrounds of mobile environment.Thus the proposed algorithm has important theoretical significance and engineering application value.