Embedding Local Quality Measures in Minutiae-Based Biometric Recognition

Degradation in data quality is still a main source of errors in the modern biometric recognition systems. However, the data quality can be embedded in the recognition methods at global and local levels to build more accurate biometric systems. Local quality measures represent the quality of local parts within a biometric sample. They are either combined into a global quality measure or directly embedded into the recognition techniques. Minutiae-based comparison is the main and the most common technique used for fingerprint recognition and high-resolution palmprint recognition in various security and forensic applications. The focus of this thesis is mainly on direct incorporation of the local quality measures into the state-of-the-art minutiae-based recognition methods, particularly those based on Minutiae Cylinder-Code (MCC). Firstly, we introduce cylinder quality measures as a new type of local quality measures associated with the local minutiae descriptors. Then, we propose several methods for incorporating such local quality measures into the biometric systems, in order to improve their recognition performance. Among them is a novel and efficient quality-based consolidation method for embedding minutiae quality and cylinder quality measures in MCC based comparison methods. We also propose a supervised embedding method based on a binary classification model, which requires labeled minutiae for training. Finally, we apply a variant of the proposed consolidation method for the challenging case of latent fingerprint and palmprint identification with embedded subjective and objective minutiae quality.