Generalized texture representation and metric

Abstract

A new approach to texture representation and discrimination is proposed. The representation is devised to reflect various perceptual aspects of image texture by retaining as much as possible the extracted texture information. Information thus extracted is then organized as a set of line or circular frequency diagrams for various texture features at different resolution levels. Based upon the properties of the two types of frequency diagrams, new metrics are defined to enhance the feature dissimilarities. The effectiveness of the proposed representation and metric for texture analysis is also illustrated by a set of actual data.