Curvature-based matching approach for automatic fingerprint identification

The performance of automatic fingerprint identification/verification systems depends on three factors: (1) the enhancement process efficiency, especially for low quality fingerprint images; (2) how the maximum available features in the fingerprint can be encoded; and (3) how much precisely the reference frame, to which the features are aligned, can be determined. We present an approach to encoding (transform) and matching the fingerprint image. In this approach, the fingerprint image is transformed into a sequence of points in the angle-curvature domain. The transformation takes into consideration the entire fingerprint features such as fingerprint pattern, ridge curvature, relative location, numbers, and characteristics (ridge ending, bifurcation, island, spur, etc.). The matching between a current template file and the database files is based on the least square error of the Euclidean distance between corresponding points in the angle-curvature domain. The number of points, that a given fingerprint image will be transformed into, is a trade off between the required accuracy on one hand and the program running time, and the storage space required for each template file on the other hand