Siamese Network-Based Palm Print Recognition

Abstract

palm print recognition is a biometric technology used to identify individuals based on their unique comfort patterns. Identifying patterns in computer vision is a challenging and interesting problem. It is an effective and reliable method for authentication and access control. In recent years, deep learning approaches have been used for handprint recognition with very good results. We suggest in this paper, a Siamese network-based approach for handprint recognition. The proposed approach consists of two convolutional neural networks (CNNs) that share weights and are trained to extract features from images of handprints, which are then compared using a loss of variance function to determine whether the two images belong to the same person or not. Among 13,982 input images, 20% are used for testing, 80% for training, and then passing each image over one of two matching subnets (CNN) that transmit weights and parameters. So that, the extracted features become clearer and more prominent. This approach has been tested and implemented using the CASIA PalmprintV1 5502 palm print database, the CASIA Multi-Spectral PalmprintV1 7,200 palm print, and the THUPALMLAB database of 1,280 palm print using MATLAB 2022a. For 13,982 palmprint recognitions in the database, the equal error rate was 0.044, and the accuracy was 95.6% (CASIA palmprintV1, THUPALMLAB, and CASIA Multi-Spectral palmprintV1). The performance of the real-time detecting system is stable and fast enough.