Orientation Field Estimation for Noisy Fingerprint Image Enhancement

Abstract There is a growing trend towards employing automatic fingerprint identification systems (AFIS) as a viable biometric tool for identity verification. However, this technology is often defeated when attempting to process low-quality fingerprints. A popular technique is to utilize Coherence-Enhancing Diffusion (CED) first to raise fingerprint quality to a level where any recognition takes place. One of the principal functions of CED is to extract orientations by referencing neighbouring gradients. Gradient computations are noise-prone and tend to deviate away from the actual underlying orientation field. This paper proposes a cluster-based computational approach for reliable orientation field estimation based on a directional filter bank framework. The orientations present in fingerprints cluster themselves in a contagious group in the directional bands, that can be exploited towards their reliable estimation. The estimated orientation field then becomes an input data for the CED engine as a pre-computed add-on module to provide much- needed enhancement in noisy fingerprints that subsequently attains in better recognition performance for a given Automated Fingerprint Identification System (AFIS). The experimental procedures reported in this study suggest more accurate orientation results in favour of proposed method than existing techniques even when processing very noisy real-life test data. The improvement in noisy fingerprints is also reported.