HYBRID DEEPEDGENET: A NOVEL MODEL INTEGRATING TRADITIONAL EDGE DETECTION TECHNIQUES WITH DEEP CONVOLUTIONAL NEURAL NETWORKS FOR ENHANCED FINGERPRINT ANALYSIS

Abstract

Today, fingerprint recognition is the most used and reliable biometric identification and security system. However, traditional edge detection methods are not able to extract fingerprint pattern features from noisy images or complex fingerprint patterns. To tackle these challenges, we propose Hybrid DeepEdgeNet, a hybrid of conventional edge detection algorithms and deep convolutional neural networks (CNN) for fingerprint classification. We successfully integrate edge detection methods such as Roberts, Sobel and Prewitt with deep learning approach of multi scale feature extraction and CNN based feature enhancement using Hybrid DeepEdgeNet. The proposed approach is a combination of the classical approach and the deep learning approach, which can capture the detailed structure information from the fingerprint images and simultaneously learn the complicated patterns. A benchmark fingerprint dataset is used to further confirm the accuracy of the proposed model, achieving overall test accuracy of 96.897% and test loss of 0.215. In specifics, the model achieved 99.2% for ID classification, 96.9% for Gender and Hand classification, and 94.7% for Finger classification. To demonstrate the effectiveness of this model on various classification problems, these results suggest that the model’s applicable in real-world biometric systems, where accuracy and stability are crucial. This work proposes the Hybrid DeepEdgeNet, a significant improvement to the fingerprint recognition paradigm, paving the way for future enhancements and the extension of this to other biometric modalities, thereby helping to define the future of secure and accurate identity authentication systems