A data fusion approach to automated fingerprint identification systems for very large databases

This thesis presents a new architecture for designing an Automated Fingerprint Identification System (AFIS) using data fusion techniques. In the system design, a two stage hierarchical model is used: A high speed preliminary stage, followed by a high accuracy final stage. Data fusion is used to take advantage of orthogonal feature representations and multiple fingerprint records. Computationally simple matching is used to limit computational complexity. A statistical model is developed to facilitate selection of appropriate system operating points.; This thesis presents the first data fusion based AFIS design. The major contributions of this thesis are: (a) the development of a hierarchical data fusion structure for AFIS; (b) the development of a compact frequency domain-based Global feature representation for fingerprints; (c) the use of Karhunen-Loeve coefficients directly for fingerprint matching; (d) the development of data fusion rules for AFIS; (e) the development of a statistical model that allows flexible tradeoff calculations between identification performance statistics and computational cost.; The experimental results and statistical projections verify that the system exhibits the desired performance, computational speed and expected algorithmic robustness. The system was tested with a NIST standard fingerprint database consisting of a wide array of fingerprint images.