TY - JOUR
T1 - Finger vein denoising algorithm based on gradient-oriented residual structure and LBP texture loss
AU - Mou, Jiale
AU - Shen, Lei
AU - Wang, Huaxia
AU - Yao, Yudong
N1 - Publisher Copyright: © 2023, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.
PY - 2024/2
Y1 - 2024/2
N2 - Finger vein recognition is an important biometric method, and its performance is easily affected by noise. The widely used Denoising Convolutional Neural Network (DnCNN) and Fast and Flexible Denoising Network (FFDNet) tend to have smoother images after denoising, and fail to fully retain information such as vein image edges and textures. To this end, a finger vein denoising net based on gradient-oriented residual structure and Local Binary Patterns (LBP) texture loss constraints, referred to as LGR-Net is proposed. First, a gradient direction operator is introduced in the residual structure to avoid the loss of gradient direction information in the denoised vein image. Second, the local features such as the texture of the shallow layer of the image are fused with the semantic information of the deep layer through the layer-skip connection, so as to preserve the semantic and local information of the denoised vein image as much as possible. Finally, an LBP loss term is added to the loss function to enhance the recovery ability of vein texture. The simulation experiments were carried out on the laboratory dust and stripe fingervein data sets. The results show that the proposed algorithm has better visual effect, Peak Signal-to-Noise Ratio (PSNR), and recognition performance, compared with the traditional Block-Matching and 3D filtering BM3D denoising algorithm, DnCNN algorithm, FFDNet algorithm and RDUNet algorithm.
AB - Finger vein recognition is an important biometric method, and its performance is easily affected by noise. The widely used Denoising Convolutional Neural Network (DnCNN) and Fast and Flexible Denoising Network (FFDNet) tend to have smoother images after denoising, and fail to fully retain information such as vein image edges and textures. To this end, a finger vein denoising net based on gradient-oriented residual structure and Local Binary Patterns (LBP) texture loss constraints, referred to as LGR-Net is proposed. First, a gradient direction operator is introduced in the residual structure to avoid the loss of gradient direction information in the denoised vein image. Second, the local features such as the texture of the shallow layer of the image are fused with the semantic information of the deep layer through the layer-skip connection, so as to preserve the semantic and local information of the denoised vein image as much as possible. Finally, an LBP loss term is added to the loss function to enhance the recovery ability of vein texture. The simulation experiments were carried out on the laboratory dust and stripe fingervein data sets. The results show that the proposed algorithm has better visual effect, Peak Signal-to-Noise Ratio (PSNR), and recognition performance, compared with the traditional Block-Matching and 3D filtering BM3D denoising algorithm, DnCNN algorithm, FFDNet algorithm and RDUNet algorithm.
KW - Finger vein
KW - Gradient operator
KW - Image de-noising
KW - LBP texture loss
UR - http://www.scopus.com/inward/record.url?scp=85174567283&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85174567283&partnerID=8YFLogxK
U2 - 10.1007/s11760-023-02795-6
DO - 10.1007/s11760-023-02795-6
M3 - Article
SN - 1863-1703
VL - 18
SP - 943
EP - 952
JO - Signal, Image and Video Processing
JF - Signal, Image and Video Processing
IS - 1
ER -