A modified CNN architecture for noise suppression in palmprint images

Abstract Palmprint is one of the most effective methods for identity verification because it remains consistent throughout an adult’s life and exhibits a variety of distinctive features. Due to the closeness in the local features of creases and ridges, the feature extraction technique for palmprints must be able to discriminate between them based on their global characteristics. If there is noise in the palmprint image, the performance of the biometric authentication or identification model suffers. Modern, cutting-edge techniques can work reasonably well, but they are very dependent on the quality of the palm print images being processed. The goal of this work is to present a method for noise suppression in palmprint images using a modified Convolutional Neural Network (CNN) that includes a nine-layer convolutional structure with zero-center normalization and a regression output layer to improve image quality. The proposed model removes three different types of noise from the images: Gaussian, salt and pepper, and Poisson noise. A quantitative analysis demonstrated that the proposed technique performed better for these three types of noise than for speckle noise. The results also show that the suggested method outperforms state-of-the-art palmprint denoising methods.