Automatic fingerprint verification systems use ridge flow patterns and general morphological information for broad classification, and minutiae information for verification. With the availability of high resolution fingerprint sensors, it is now feasible to capture more intricate features such as ridges, pores, permanent scars, and incipient ridges. These fine details are characterized as level-3 features and play an important role in matching and improving the verification accuracy. The main objective of this research is to develop a fast and accurate quality induced multiclassifier fingerprint verification algorithm that incorporates both level-2 and level-3 features. A quality assessment algorithm is developed that uses Redundant Discrete Wavelet Transform to extract edge, noise and smoothness information and encodes into a quality vector. The feature extraction algorithm first registers the two fingerprint images using a two-stage registration process. In the first stage, Taylor series based image transformation is used to perform coarse registration while in the second stage, thin plate spline transformation is used for fine registration. Then, a fast Mumford-Shah curve evolution algorithm is used to extract four level-3 features namely, pores, ridge contours, dots, and incipient ridges. Gallery and probe features are matched using Mahalanobis distance measure.Correlation analysis suggests that level-2 and level-3 features can be combined to improve the verification performance. Therefore, we propose five different match score fusion algorithms to combine the match scores obtained from level-2 and level-3 features. The first algorithm uses Delaunay triangulation to obtain invariant features related to level-2 and level-3 information and then combines them to generate a fused match score. The next three match score fusion algorithms utilize different techniques in information fusion namely, density based approach, classifier learning, and belief models. Experimental results show that the proposed evidence theoretic sum rule algorithm yields good performance under ideal conditions. However, if the match scores provide conflicting decisions, more sophisticated techniques are required. Belief models based fusion algorithms are ad-hoc in nature and learning algorithms require representative training dataset for correct classification. To address the limitations of these three techniques, we propose a sequential fusion algorithm which combines the learning theory and belief model with the statistical approach. The sequential fusion algorithm yields good verification performance at the cost of computational complexity. To optimize both verification accuracy and computational complexity, we introduce the concept of unification framework that takes into account the variability in image quality, and the characteristics of level-2 and level-3 features to select the most appropriate fusion algorithm. Experimental results on a high resolution fingerprint database show the effectiveness of the proposed algorithms.We further propose a novel biometric watermarking algorithm to embed the level-2 and level-3 features in the face image of the same individual for increased robustness, security, and accuracy. The proposed watermarking algorithm first computes the embedding capacity in the face image using edge and corner phase congruency method. Embedding and extraction of fingerprint features is based on redundant discrete wavelet transformation. Moreover, the proposed watermarking algorithm uses adaptive user-specific watermarking parameters for improved performance. Experiments on the face-fingerprint database show that the proposed watermarking algorithm is robust to different frequency and geometric attacks, thereby securing the biometric data against tampering.
展开▼
