Research On Personal Identification Method Based On Finger-kunckle-print

Currently iin the network information society,many occasions require a true and reliable identification of people's identity.At present,biometric identification technology is recognized as the most effective identification method.As a kind of biometric recognition,finger-knuckle-print(FKP)recognition has urnique advantages and broad application prospects.Compared with other mature biometric recognition methods,FKP recognition has developed rapidly.But due to its late presentation and insufficient research,there are still some problems to be solved.Firstly,traditional recognition methods have redundant features and low recognition accuracy.Secondly,in order to achieve practical application in identity recognition,it is necessary to achieve fast recognition,while the related research on recognition speed is relatively deficient.In this paper,based on PolyU-FKP database,aiming at the above problems,traditional algorithms and deep learning methods are studied.The specific research contents are as follows:(1)For the problem that the line structure of the FKP ima"ge is unclear and the contrast is low,the contrast limited adaptive histogram equalization is used to enhance the image texture and improve the contrast;(2)The FKP recognition is carried out by the traditional algorithm,the Sift features of the FKP image are extracted for matching,and phase correlation algorithm combining LBP features is proposed.The experimental results show that the feature extracted by the traditional algorithm is redundant and the recognition accuracy is low.The improved method improves the recognition accuracy;(3)For the problem that the traditional algorithm features large redundancy and low recognition rate,the deep belief network is used to extract feature and classify the FKP image.The deep learning method can effectively solve the feature redundancy problem.The LBP feature is extracted from the enhanced image,and as the input of the network.The influence of local transformation on the features learned by the deep belief network is reduced,such as translation and rotation with uneven illumination and small amplitude.After changing the input to the LBP feature,the recognition speed is improved by 10 times compared to the direct use of the original image;(4)For the problem that the recognition speed is slow and the feature representation is less robust to the local transformation of the input,the convolutional neural network and the deep belief network are combined to form a convolutional deep belief network,and the LBP feature is used as the input.The recognition speed increased by 24 times compared to the deep belief network;(5)Based on the MATLAB platform,an identification system based on the FKP image was developed.The system can select a single FKP image in the folder,perform contrast limited adaptive histogram equalization,extract LBP features,and use the deep belief network(DBN)and the convolutional deep belief network(CDBN)to identify FKP image and obtain category and recognition time.The recognition accuracy and the time required for the left index,the left middle,the right index,and the right middle and the whole test set can be obtained.