Research On Finger Vein Image Quality Evaluation And Recognition Algorithm

Finger vein recognition technology uses finger vein information for identification,which has the advantages of uniqueness and stability.In practical applications,vein information takes images as carriers.Therefore,the quality of the finger vein image is one of the critical factors in determining the performance of the finger vein recognition system.Evaluating the quality of finger vein images and eliminating low-quality images will help improve the accuracy and robustness of the recognition system.Although the existing finger vein image quality assessment methods can effectively evaluate the image quality,there are still some shortcomings.With the rise of deep learning technology,many researchers have applied it to image quality assessment and achieved good results.This paper studies the quality assessment method of finger vein images based on deep learning and mainly does the following work.(1)A finger vein image quality assessment algorithm based on data uncertainty is proposed.The existing finger vein image quality assessment algorithms are not objective enough and lack universality.Manually designed indicators are biased toward subjective judgments.And a single indicator is too one-sided to measure the quality of the image accurately.Aiming at these problems,this paper proposes a new image quality assessment algorithm.The algorithm analyzes and applies the uncertainty in the finger vein image to evaluate the image quality.Firstly,the network model is used to model the random embedding with Gaussian distribution,representing the uncertainty in the finger vein image.Secondly,the uncertainty estimation of the image is obtained by classification learning.Then,the uncertainty estimation of the image is fused with other indicators to calculate the image quality score.Finally,experiments are carried out on public data sets and self-built data sets to verify the effectiveness of the proposed algorithm.(2)A finger vein image quality assessment algorithm is proposed based on a Bayesian neural network.Existing neural network generally use single-point estimation for weights.From the perspective of probability theory,this method is unreasonable for image classification,and the extracted features are not comprehensive enough.Bayesian neural network can not only output the predicted value but also the confidence of the predicted value.This paper proposes a new image quality assessment algorithm using the characteristics of Bayesian neural network.First,the algorithm obtains the predicted distribution of the output through a Bayesian neural network.Second,the instability of the image output is obtained by analyzing the predicted distribution.Then,the instability of the image output is fused with other indicators to get the quality score of the image.Finally,through multiple sets of experiments,the evaluation performance of the algorithm in this paper is proved.