Optimal keys for image database indexing

Abstract

This paper examines the problem of efficient indexing of large image databases using the theory of optimal keys. The methods based on optimal keys are compared empirically with a texture classification method and template matching for benchmarking purposes in the Leiden WWW color image database and the 19th century portrait database. The different indexing methods are compared and evaluated in the problem space of finding copies of corrupted images. Real world noise is present in the form of print-scanner noise and general image degradation.