Research And Implementation Of Finger Vein Image Quality Assessment Based On Deep Learning

In today's intelligent society,biometric identification has gradually been widely used in personal identification.Finger vein technology has become an important part of biometric identification and has been paid much attention by researchers.Based on the principle that the finger vein will form dark vein texture in the image taken by infrared camera,finger vein recognition algorithm is carried out through the captured image.Therefore,the quality of the obtained finger vein image is the key factor affecting the accuracy of the recognition algorithm.Combined with the method of deep learning,this paper mainly studies the issue of finger vein image quality and proposes corresponding solution.The results are as follows:(1)In this paper,a manually annotated finger vein feature dataset is contributed.And on this basis,a network model(FV-UNet)that can extract the features of finger vein is proposed.This is the first finger vein feature extraction algorithm based on deep learning through manually labeled finger vein images.(2)The methods of finger vein image quality annotation in the form of discrete value and continuous value are proposed.In this paper,the high-quality and low-quality images are marked in the form of discrete values based on the method of recognition accuracy.Then,the problem of data imbalance is solved through a combination of various methods.On the basis of discrete value labeling,a continuous value labeling method for finger vein image quality is proposed(3)A model of finger vein image quality assessment based on deep learning is proposed.Since the image quality of finger vein should largely depend on the venous features contained in the image,the design of this model is extended from the model for extracting venous features.The validity of the model is proved by a low-quality image filtering experiment on the recognition system.(4)A finger vein image quality assessment system is designed and implemented.This system integrates the feature extraction and quality evaluation of finger vein image researched in this paper,to simulate the application in real scene.