Single Footprint Based Person Identification And Verification

Biometric identification technology has better security,reliability and effectiveness than the traditional password or ID card identification method.Studies have shown that human footprints are as important as other biometric features,such as face and fingerprint,and the information on footprint morphology of every person is unique.Therefore,footprint based personal identification is a promising and alternative biometric method.The existing works can be divided into three categories according to their different input forms:standing foot pair based methods,walking footprint sequence based methods and single footprint based methods.Methods based on standing foot pair or walking footprint sequence have certain limitations on convenience and stability of data acquisition.Single footprint based biometric recognition requires a simple equipment and low cost,and it can realize personal identification and verification,so it has much wider applicability and great practical value.The main purpose of this thesis is to study personal identification and verification algorithm based on single footprint.The works are as follows:1)Two kinds of foot representation method are proposed.According to the structural characteristics and local shape relationship of the foot,a weighted local structure is proposed to represent the geometric shape of a single footprint.Since each person's walking and standing habits have obvious differences,a regional weighted pressure direction strength feature of the single footprint is extracted.2)A feature discrimination and mutual information combination based feature fusion algorithm is proposed.Feature discrimination and mutual information are used to compute feature fusion weight,which makes the performance of single footprint identification and verification system better.3)A selective multi-classifier decision fusion algorithm is proposed.The method uses the accuracy of the classifier and the difference between the classifiers to select the classifier with good performance and large difference.At the same time,to make full use of the advantages of different classifiers,the multi-scale voting method is adopted to improve the single footprint identification performance.4)To evaluate the performance of single footprint recognition algorithm,a MUR2RV2 footprint database is created,which contains 4748 single footprint images from 1115 volunteers.The database contains three subsets of MUR2RV2-I,MUR2RV2-V and MUR2RV2-C,and they can be used to evaluate the performance of identification,verification and permanence of the features.Three images are randomly selected as gallery set images from each category of the MUR2RV2-I dataset,and the remaining 1115 images are used as probe images.The average recognition accuracy of the four experiments is taken as final recognition accuracy.The experimental results have shown that the proposed algorithm is superior to the existing algorithms,and the recognition accuracy has reached 97.13%.13380 positive pairs and 19783760 negative pairs from MUR2RV2-V were used to evaluate the verification performance.When the FAR is 0.1%,the FRR is 5.97%.EER is 1.67%.The proposed algorithm has better verification performance than the existing algorithms.Experimental results on MUR2RV2-C have shown the identification and verification performances don't change significantly with the change of time and equipment,which proves that the proposed footprint representation features have good permanence.