Composite Artificial Intelligence

8Fingerprint Registration and Matching Based on Improved Convolutional Neural Network

Lakshmanan B.^*, Selvakumar B., Kasthuri K., Nivashini S. and Swetha R.

Department of Computer Science and Engineering Mepco Schlenk Engineering College, Sivakasi, Tamil Nadu, India

Abstract

The problem of latent fingerprint matching is significant yet unresolved. As an essential step in fingerprint matching, registration of fingerprints has an important effect on recognition performance. Existing techniques for registering latent fingerprints mostly depend on identifying correspondences between minute details, and they will almost surely fall short if insufficient minutiae are retrieved due to a tiny fingerprint regions or noisy images. This work, represented an improved method for latent fingerprint registration based on human fingerprint patch alignment and matching; it assesses the spatial transformation between two fingerprints. The presented work is used to enhance recognition performance of the latent fingerprint with various challenging conditions like tiny regions, poor image quality, and severe distortion. In this proposed method, deep learning-based convolution neural networks (CNN) is adopted for feature extraction, descriptor extraction, and matching similarity score. In this presented work, we skip the minute detail extraction stage and use evenly sampled locations as critical points when we have a matched pair of fingerprints. Each pair of sampling points is then compared using ...