FvMamba: Super-Resolution Reconstruction Technology of Finger Vein Based on UNet-Like Mamba

To address the issue of elevated false detection rates in finger vein recognition, which arise from blurred images due to external environmental factors and hardware limitations during image acquisition, we propose a U-Net-based super-pixel image reconstruction method for finger veins utilizing Mamba blocks. This approach integrates Mamba blocks into the decoder to effectively capture long-range dependencies within the images, enhances the model's receptive fields, and better preserves the linear texture features characteristic of finger vein images. Experimental results demonstrate that our proposed method significantly improves the quality of finger vein images. When compared with SRCNN, EDSR, and FSRCNN methods on the FV-USM dataset, our approach exhibits superior performance in feature extraction as well as higher quality reconstructed images, achieving improved PSNR and SSIM metrics relative to other techniques.