Gunet: a novel and efficient low-illumination palmprint image enhancement method

Palmprint has a high application prospect due to the stability, uniqueness, difficulty of reproduction, easy acquisition and high user acceptance of its own texture characteristics. However, palmprint images acquired at long distances and in low light conditions lose a amount of significant palmprint texture features, which in turn will affect the accuracy of palmprint recognition. Combining traditional methods with deep learning methods, this paper proposes a low-illumination palmprint image enhancement method based on wavelet transform and GUNet (insertion of Integrated Attention Gate at the UNet jump connection). Firstly, the input image is decomposed by wavelet transformation to obtain the decomposition result of the original image, that is the high-frequency image and the low-frequency image. Secondly, the high-frequency image is enhanced by the GUNet neural network to enhance the palmprint texture, and low-frequency images use weighted averaging to smooth low-frequency information other than palm texture information. After that, the palmprint image with a clear palmprint texture is obtained by the wavelet inverse transformation. Numerous experiments are carried out on palmprint databases such as Idiap, CASIA, IITD and a laboratory self-collection. Experimental results show that the proposed approach has higher Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM) and Visual Information Fidelity (VIF) indicators than some existing methods, which suggest palmprint image with low-illumination can be significantly enhanced.