2S-CFE: a two-stage contourlet-domain framework for contactless fingerprint enhancement

Abstract Contactless fingerprint recognition offers significant advantages over contact-based sensing in terms of hygiene, faster acquisition, and reduced pressure-induced distortion. Nevertheless, contactless images typically exhibit low ridge-valley contrast, uneven illumination, perspective variation, and unreliable local ridge-frequency estimation, all of which degrade minutiae extraction and matching. Therefore, we propose a novel two-stage contactless fingerprint enhancement framework. The first stage performs contourlet-domain Block Matching and 3D Filtering (BM3D) denoising, while the second stage estimates local ridge orientation and frequency, applies adaptive Gabor filtering, and performs region-quality-aware minutiae selection, encoding, and matching. Experiments conducted using the Hong Kong Polytechnic University (PolyU) contactless fingerprint database prove the effectiveness of the proposed enhancement technique, with a consistent decrease in Equal Error Rate (EER) across both contactless-to-contact-based and contactless-to-contactless matching scenarios. In the contactless-to-contact-based scenario, the Equal Error Rate decreases from 49.18 to 21.97% for M1, from 48.86 to 26.73% for M2, from 38.90 to 11.22% for M3, and from 24.76 to 17.50% for M4. In the contactless-to-contactless scenario, the Equal Error Rate decreases from 42.91 to 13.10% for M1, from 43.25 to 12.74% for M2, from 41.90 to 11.90% for M3, and from 31.00 to 9.05% for M4. These results demonstrate that the proposed enhancement framework consistently improves the performance of all evaluated matchers in both matching scenarios.