Feature Similarity-Based Finger Vein Image Quality Assessment Method

Image quality has a large impact on the performance of finger vein recognition systems. Image quality assessment algorithms can be used to identify and reject low-quality images timely and accurately at the time of image acquisition. Existing assessment methods for finger vein images either rely on the subjective judgment of experts or only make a simple distinction between high and low quality, which are not practical enough. In this paper, a feature similarity-based finger vein image quality assessment method (FS-FVIQA) is proposed to address these problems. Firstly, the images are pre-processed with histogram equalization, and the ResNet50 network is used to train the histogram equalized images to obtain the image feature vector. Then the Wasserstein distance score of the intra-class similarity distribution and inter-class similarity distribution of the images, and the cosine matching score are calculated. Finally, the two scores are weighted and fused to obtain the image quality. Simulation tests are carried out on a publicly available dataset. Simulation results show that the proposed algorithm can effectively reduce the equal error rate.