Finger Vein Recognition Algorithm Based On Low Image Quality

Biometric identification is defined as utilizing unique physiological characteristics of the human body for identity authentication.The common physiological features with uniqueness and stability are fingerprints,faces,iris and veins etc.The finger vein is located inside the finger and is a kind of living feature with high security and anti-forgery,and its image acquisition equipment is low cost and convenient to use.So finger vein recognition is becoming the hot direction of biometric recognition.At present,the products integrated with finger vein recognition include ATM,gate,entrance guard etc,and have a certain share in the biometric recognition market.Therefore,finger vein recognition is a technology with broad development prospect and great research value.The main problems to be solved in the large-scale application of finger vein recognition are the low recognition performance and accuracy of low-quality images.There are two reasons for this,and one of the reason is that there are few veins in a certain proportion of finger veins.The collected image is a fuzzy low-quality image with less information available in the venous region and poor recognition performance.The other reason is that the acquisition area of the equipment is polluted by the long term use,and then the image collected is a low quality image containing fixed noise.When both images have fixed noise,the risk of false recognition will increase rapidly.Therefore,the finger vein recognition technology is focused in this paper,aiming at the two problems mentioned above,in order to improve the performance of low-quality blurred image recognition and control the risk of false recognition after long-term use of equipment.The specific research contents in this paper are as follows:An algorithm of finger vein mage feature extraction based on multi-scale,multi-directional two-dimensional Gauss template curvature calculation is proposed.The algorithm first calculates the curvature of the image and then extracts three features from an image according to the curvature value:background curvature gray feature,vein curvature gray scale feature and curvature thin line feature.Compared with the traditional algorithm using directional filter to extract thin line features,the curvature thin line feature extracted in this paper has better recognition performance,which shows that the computing curvature has better effectiveness in image enhancement.A finger vein recognition algorithm based on the fusion of three kinds of feature thresholds is proposed.The threshold of which the three features extracted in this paper are identified separately is weighted and fused,and the recognition algorithm is designed based on the fusion threshold mentioned above.The performance of the proposed fusion recognition algorithm is proved from both theoretical and experimental aspects to be superior to that of the three features separately,and the three features extracted in this paper contain the information of vein region and background region.Compared with the traditional algorithm which only extracts the feature of vein region,especially the targeted performance improvement of low-quality blurred images,so the performance of fusion recognition is better than that of traditional algorithms those only recognizing with vein region information.A distribution fitting method based on the minimum error criterion of intersecting region is proposed.The normal fitting of threshold Data of Three Features Recognized separately is made by using the proposed Distribution Fitting Method.The best fusion weight calculated based on fitting parameters coincides with the actual optimal fusion weight,which shows that the proposed distribution fitting method is more suitable to the actual situation than the conventional data fitting method.In particular,it solves the problem that the application of threshold fusion recognition algorithm is limited in the case of small data samples.A fixed noise detection algorithm based on the principle of random distribution of veins is proposed.The randomly distributed venous area in the image is smoothed off by the superposition average,the fixed noise part is preserved,and the contamination degree of the current equipment is determined by detecting the area of the fixed noise part.When the risk of false recognition is increased,users are reminded to clean the equipment,to ensure the long-term availability of the equipment,and reduce the risk of fraud.