Automated Biometrical Fingerprint Recognition Scheme using Synthesized Images

Abstract

The evolution of digitization has engulfed various methods of forensic sciences, such as fingerprint detection, recognition or recovery of partial prints. Prior to computerization, huge fingerprint repositories were manually maintained and involve humans for classification. But the advent of artificial intelligence-based tools performs the fingerprint recognition much faster and easier. Therefore, this study proposed CNN-based deep learning technique to extract effective features from ridges and valleys of skin impression for accurate recognition. The experimental work is based on FVC2020 dataset to train and test the proposed model. Moreover, ResNet50 framework is also tested on this dataset, and results shows that proposed model achieved an accuracy of 81.25%, whereas ResNet50 attained 79.25% accuracy. Furthermore, the incorporation of convolutional auto-encoder (CAE) based model for enhancing the dataset by generating synthetic fingerprint images, improved the recognition accuracy of proposed CNN-based model to 86.0%.