3D Palmprint Recognition Based On Multiple Features

In recent years,biometric technology has received more and more research and attention,it can provide us with an effective and reliable personal identification method.As a relatively new biological feature,palmprint identification has a wide range of applications due to its strong discrimination,user-friendliness,and rich features.Although 2D palmprint has the advantages of easy acquisition and low cost,but 2D palmprint images are sensitive to light,angle,and surface dirt and are possibly deceived by fake palms.The 3D palmprint recognition technology has overcome the bottleneck of the 2D palmprint recognition technology and has become a research hotspot at home and abroad.3D palmprints are no longer limited by light and angle,but the deformation of palms and external noise are still inevitable.Therefore,to discover the inherent features of high discrimination and stability and to design the feature representations that are robust to deformation and noise are key problems for 3D palmprint recognition.Based on the in-depth analysis of 3D palmprint features,this paper studies the 3D palmprint recognition algorithm based on multiple features.The main research work and innovations of this paper are as follows:1.A 3D palmprint recognition method based on multi-level palmprint features called SSM and CR model is proposed:1)For surface type(ST),points near the threshold are easily misjudged due to deformation and noise interference,which leads to the uncertainty of geometric feature expression.In order to solve this problem,this paper proposes to use the shape index coding called SIC to jointly characterize the geometric features of 3D palmprints.Because of the complementarity between SIC and ST,the stability of 3D palmprint geometric feature expression can be improved.2)A multi-scale modified competitive coding called MSMCC is proposed.Compared with a single-scale Gabor filter,MSMCC can obtain more orientation information.Compared with traditional competitive coding,MSMCC divides the palmprint into flat regions and normal regions.Then,the points on the flat areas are uniformly coded in MSMCC,which can achieve a good balance between the discrimination and stability.Related experiments indicate that MSMCC can better express the orientational characteristics of palmprints.3)The ST,SIC and MSMCC(SSM)are fused at the decision-making level.The performance of the four classification models(SVM,Google Net,sparse representation,and collaborative representation)with the SSM features is compared in this paper.Considering the recognition rate and computing efficiency,finally a multi-dictionary CR-based model is choosed,which further improves the accuracy and robustness of the algorithm.The proposed method can maintain the low computational complexity with a high recognition accuracy.2.A 3D palmprint recognition method based on multi-points voting local ternary pattern is proposed:1)Because palmprint is a non-rigid object,it is susceptible to deformation and noise when being collected.The related 3D palmprint identification technology fails to make full use of local neighboring information when extracting local descriptors.When faced with such local deformation,the multi-points voting local ternary pattern which is called MPVTP can maintain high stability and accuracy because it utilizes the information of multiple neighbouring points,and is particularly suitable for palmprint identification.2)Combine MPVTP with ST,SIC and competitive coding to form a complete multi-points voting ternary representation caleed CMPVTR of a palm and apply it to 3D palmprint identification.Internal comparison experiments were performed on the levels and voting point amounts of the MPVTP.Also,the combination experiments proved the effectiveness of combined MPVTPs.On the Poly U3 D palmprint database,the Rank-1 recognition rate of our method is 99.59% and the EER is 2.32%.3.MPVTP is a feature descriptor that is robust to local viriations.It can combine with existing features to further extract more discriminative and stable feature representations.In this paper,cross-contrast experiments are performed between the algrithms of Chapter2 and Chapter3.The experimental results show that after combining the typical features with MPVTP,the Rank-1accuracy is increased by 2.46%,and the EER is reduced by 0.55%,but CMPVTR needs longer feature extraction time and palmprint matching time.Therefore,the two 3D palmprint recognition algorithms proposed in this paper have their own advantages and disadvantages,they can meet the needs of different application scenarios.