AUTOMATIC SHOEPRINT IMAGE RETRIEVAL SYSTEMS: A COMPARATIVE STUDY

摘要

Shoeprints are recently of great interest to police and forensic scientists. Researchers examine how police's search into crime scenes could be enhanced through matching suspects shoeprints using automated computer systems. In this paper we attempt to study and compare two shape descriptors which have been adopted for shoeprint matching, these are: Hu's moment invariants (HMI) and the combined Topological and Pattern Spectra (TPS) descriptors. Shape descriptors in the Content-based Image Retrieval (CBIR) should satisfy several properties such as compact representation, robustness, retrieval performance and computation complexity. A database of 500 'clean' shoeprints is used to evaluate the performance of the techniques. Five test databases are generated, each with 2500 images degraded with Gaussian noise. Retrieval results demonstrate the comparison between the two methods against these properties.