Research On Finger Vein Recognition Algorithm Based On Generalized Label Model

In biometrics,finger vein recognition have received much attention from the society because of it's high accuracy,high reliability and living body recognition.With the growing demand for information security,finger vein recognition has also become widespread.At present,finger vein recognition mainly focuses on the extraction of region of interest,image feature enhancement and high reliability recognition algorithms,but there are still many problems such as low region extraction accuracy,weak image optimization ability and less recognition robustness.Therefore,based on the above steps and problems,the research content and contributions of this paper are as follows:(1)In order to enhance the adaptive ability and accuracy of finger vein region extraction,the finger vein image semantic segmentation method is adopted to understand the image through mask region convolutional neural network and accurately segments the finger vein,getting finger vein segmentation mask,recognition area,and class probability results.The largest rectangular region is extracted from the irregular segmentation mask based on the improved bidirectional ergodic central diffusion method,because it can facilitate subsequent recognition.The experimental results show that the effective extraction rate of this method in the SDUMLA-HMT data set reaches 100%.(2)In order to better eliminate finger vein image noise and brightness interference,this paper proposes an improved denoising auto-encoder based on residual learning and parallel asymmetric convolution,which enhances the image quality by enhancing the reconstruction of hidden layer features.On this basis,the sub-task network algorithm is established to achieve denoising and adaptive balance brightness based on the statistical characteristics of grayscale image.The experimental results show that the proposed algorithm achieves good results in both the finger vein image and the natural image.Compared with the traditional algorithm,the original image can be restored more clearly,which has certain adaptability and mobility.(3)The unknown label and the passive matching of the finger veins will cause serious interference in recognition.This paper uses a deeper convolutional neural network to improve the finger vein feature extraction ability.After the known label finger vein recognition is stable,the unknown label will be generalized into a class to reduce interference.Then,based on the classification results,a label-based verification threshold adaptive acquisition algorithm is proposed,which makes the classification identification and verification uniform.Based on the previous good region extraction and image enhancement,the recognition algorithm get lower error rates of 1.436% and 2.21% in the LNTU-FVD and SDUMLA-HMT finger vein database respectively.The paper has 42 pictures,21 tables,and 76 references.