Finger Vein Recognition Method Based On Multi-Feature Fusion

With the advent of the information age and the continuous development of computer technology,there are higher requirements for the accuracy and security of identity recognition.The identity markers(such as ID cards,keys,passwords,etc.)used in the traditional identification technology have security risks that are easily lost or forgotten,which leads to the risk of being unable to authenticate or being impersonated.Biometric identification uses human physiological characteristics or behavioral characteristics for identity authentication.It has the characteristics of invariance,uniqueness,and difficulty in losing and stealing.It is more secure in security and more convenient in use.In recent years,biometric identification technology has developed rapidly,and biometric features such as fingerprint and face have been widely used in the identification field.Compared with fingerprint recognition and face recognition,finger vein recognition has some unique advantages.For example,during the process of collecting finger vein images,users do not need to contact with the collection device,which is hygienic and easy to accept by users.Finger vein belongs to the internal characteristics of fingers,so it is difficult to be stolen.Finger vein recognition has the characteristics of live authentication and high safety.Therefore,finger vein recognition technology has received attention from researchers,and a number of research results have been achieved.Most of the existing methods use a single image feature for finger vein retrieval or recognition.Aiming at the problem of poor discrimination of single image feature,this paper studies the improvement of finger vein recognition performance(i.e,efficiency and accuracy)based on multi-feature fusion.Studies have shown that multi-feature fusion is an effective method to improve recognition performance.Multi-feature fusion can more effectively utilize the complementarity of different features to get more discriminative fusion features,and therefore the better recognition performance can be achieved.Therefore,this paper applies multi-feature fusion for finger vein image retrieval and recognition,making full use of the correlation and complementarity of multiple features to further improve the recognition efficiency and accuracy.The main work of this paper includes :(1)In order to improve the efficiency of finger vein recognition,a finger vein image retrieval method based on multi-feature fusion is proposed.In the proposed method,texture feature,orientation map,and vein backbone are extracted from finger vein images,and then texture histogram and orientation histogram are defined by an improved multi-texton histogram method.Two kinds of histograms are concatenated for finger vein image retrieval.Moreover,for exploring the spatial distribution of finger vein,one image is partitioned into multiple blocks,and multi-feature fusion is performed in block-wise.The proposed method is tested on two public finger vein databases.The experimental results show that the proposed method powerfully fused multiple features and improved the retrieval performance.(2)In order to improve the accuracy of finger vein recognition,a finger vein image recognition method based on multi-feature fusion is proposed.Firstly,the Gabor filters are used to extract direction features,texture features,and vein patterns from finger vein images.Then,the dictionary learning method is extended to the consistent multi-feature dictionary learning method.In this method,dictionaries are learned from different features,and the same coding vector is computed for one image from different dictionaries by consistent dictionary learning.Lastly,to make the coding coefficient more discriminative,we add one local constraint to represent one image mainly by its similar dictionary atoms.The proposed method is tested on the self-built finger vein database and two synthetic finger vein databases.The experimental results demonstrate the effectiveness of the proposed method for finger vein recognition.