Methods for iris classification and macro feature detection.

摘要

This work deals with two distinct aspects of iris-based biometric systems: iris classification and macro-feature detection. Iris classification will benefit identification systems where the query image has to be compared against all identities in the database. By preclassifying the query image based on its texture, this comparison is executed only against those irises that are from the same class as the query image. In the proposed classification method, the normalized iris is tessellated into overlapping rectangular blocks and textural features are extracted from each block. A clustering scheme is used to generate multiple classes of irises based on the extracted features. A minimum distance classifier is then used to assign the query iris to a particular class. The use of multiple blocks with decision level fusion in the classification process is observed to enhance the accuracy of the method.;Most iris-based systems use the global and local texture information of the iris to perform matching. In order to exploit the anatomical structures within the iris during the matching stage, two methods to detect the macro-features of the iris in multi-spectral images are proposed. These macro-features typically correspond to "anomalies" in pigmentation and structure within the iris. The first method uses the edge-flow technique to localize these features. The second technique uses the SIFT (Scale Invariant Feature Transform) operator to detect discontinuities in the image. Preliminary results show that detection of these macro features is a difficult problem owing to the richness and variability in iris color and texture. Thus a large number of spurious features are detected by both the methods suggesting the need for designing more sophisticated algorithms. However the ability of the SIFT operator to match partial iris images is demonstrated thereby indicating the potential of this scheme to be used for macro-feature detection.