Improving Fingerprint Recognition With CNN

Abstract

A deep learning-based fingerprint recognition system is developed to match contactless fingerprint inputs with stored contact-based fingerprint records. Using a lightweight yet powerful MobileNetV2 architecture, the system efficiently extracts deep features from fingerprint images while maintaining high accuracy and fast performance. The model is trained with a distance-aware loss function and incorporates ridge-based features to improve matching reliability between the two modalities. When a user uploads a contactless fingerprint image, the system preprocesses it, generates a feature embedding, and compares it with a database of contact-based fingerprints to identify the best match.

The matched person ID is then displayed through an attractive and interactive user interface designed for ease of use. This solution avoids physical contact with sensors, making it suitable for touch-free, hygienic authentication in public or mobile environments. The integration of MobileNetV2 makes the system highly scalable and lightweight, suitable for deployment on both desktops and embedded platforms. Overall, this project demonstrates the practical application of deep learning in bridging modality gaps in fingerprint biometrics, offering a robust and future-ready identity verification system.