Method for image informational properties exploitation in pattern recognition environment

Abstract

The problem of finding optimal image transformation according to informational image properties is considered. During the research, image informational properties that distinguish images from all other recognition objects were investigated. The notion of image equivalence was introduced. Notion of image equivalence as identity of images with respect to some transformation set was explored. The proposed equivalence definition allows decomposition of image set into subsets of a certain type and establishes correspondence between image equivalence classes and some subsets of operations. On this basis, the method for selecting efficient image recognition algorithm according to image informational nature was elaborated. The proposed method provides a possibility of improving efficiency of selecting image analysis algorithms and automation (partial or complete) of image processing. It allows taking into consideration internal information that image conveys, syntactic and semantic. © Springer-Verlag 2003.