A texture-based distance measure for classification

Abstract

A distance measure based on a new representation scheme of texture images is presented. The new representation scheme captures the structural and statistical properties of a homogeneous region of texture. Each region is represented by a set of feature frequency matrices (FFM) which gives the frequencies of occurrence of joint feature events. Feature events are extracted by operators defined by users and/or applications. The representation is further refined by applying a hierarchical maximum entropy partitioning scheme to the FFM. The proposed distance measure is a weighted function of the partitioned FFM. The novelty of this measure lies in the process of determining the weights. In a classification experiment, we shall demonstrate the efficacy of the distance measure.