Robust Fingerprint Minutiae Extraction and Matching Using Fully Connected Deep Convolutional Neural Network and Improved SIFT
Saif O. Husain, R Archana Reddy, Piyush Kumar Pareek, Srinivasan Jagannathan, M. Jyothi · 2024
Fingerprint is a most well-known biometric based authentication system that gives a unique identity for each person. In this paper, an authentication framework to enhance the Fingerprint Minutiae Extraction and Matching (FMEM) technique is proposed. A Fully Connected Deep Convolutional Neural Network with Improved Scale-Invariant Feature Transform (FCDCNN-ISIFT) is proposed for feature extraction and feature matching. Initially, the fingerprint images are collected from Fingerprint Verification Competition 2004 (FVC2004) which are first pre-processed using Histogram Equalization (HE) and Normalization. The images are segmented and enhanced by estimating ridge orientation and ridge frequency, and performing Gabor filtering, binarization and thinning. The fingerprint feature extraction and matching test is done using FCDCNN-ISIFT. The FCDCNN is built for detecting minutiae and the ISIFT is applied to the high contrast images to enhance the quality of latent fingerprints. Finally, the loss function is measured to determine the minutiae detection loss. The experimental results show better accuracy of 99.41%, f1-score of 97.73%, and Equal Error Rate (EER) of 1.75% when compared to results of existing approaches like Contrast Limited Adaptive Histogram Equalization - Random Sample Consensus (CLAHE - RANSAC) and Deep Convolutional Neural Network (DCNN).