A Study On Palmprint Recognition Based On Local Feature Descriptors

Palmprint recognition is a new biometric identification technology.Low-resolution palmprint recognition has the advantages of high recognition accuracy,low-cost acquisition equipment,and good user acceptance.The technology is currently more at the theoretical research stage.However,the advantages of low-resolution palmprint recognition give it great potential for future applications and commercialization.Distortion,noise and blurring of palmprints are some of the key factors affecting low-resolution palmprint recognition.Among them,however,blurring is less common,especially in the public palmprint databases currently available.Therefore,we leave the factor of blurring out of the scope of this study and focus on distortion and noise.Palmprint distortion refers to the geometric deformation of palmprint images due to different palm placement positions and stretching degrees during palmprint acquisition.It includes translation,rotation,scaling and other nonlinear deformation.The presence of distortion will cause user identification failure.Current palmprint recognition algorithms have not studied this problem well.In order to solve this problem,thesis presents a new palmprint matching algorithm based on the Harris corner detector and SIFT operator.The main work of this paper is as follows:Firstly,prior to the extraction of local feature points,the MFRAT encoding method is used to filter the noise-contaminated palmprint texture features.This operation not only improves the robustness of the algorithm to noise,but also improves the contrast of palmprint lines and deepens palmprint texture information.Secondly,this thesis proposes to use the Harris corner detector and the SIFT operator to extract the local features distributed around palm lines and folds.These feature points contain rich palmprint texture information.The acquisition of local feature points ensures the accuracy of palmprint recognition and also reduces the computational complexity of the algorithm.Thirdly,after extracting the feature points,considering that the palmprint images may be affected by translation,rotation,etc.,the extracted feature points must have a position deviation when they are mapped to the query palmprint image.This thesis uses an image pyramid algorithm to correct palmprint feature points in a local area,such different areas will have different shape variables,which are more flexible and improve the recognition accuracy.Finally,during feature matching,this thesis proposes a statistical matching method by analyzing the offset information of the feature locations.Many classic phase correlation methods for palmprint recognition ignore the offset information.This algorithm is another way to use the offset information to identify and match palmprint images.Experiments show that the matching method can significantly improve the recognition accuracy of palmprint.Experiments are performed on the Poly U and CASIA palmprint databases.respectively.The EER(equal error rate)of the classic palmprint recognition algorithm is 0.12%,The experimental results show that the best recognition rate of the Poly U database palmprint database is 0.0105%,and the best recognition rate of the CASIA palmprint database is 0.0644%.The algorithm proposed in this study is superior to the classic palmprint recognition matching algorithm.