Scale-Space Texture Analysis

In this paper we propose a technique for classifying images by modelling features extracted at different scales. Specifically, we use texture measures derived from Pap smear cell nuclei images using grey level Co-occurrence Matrix (GLCM). For a texture feature extracted from GLCM at a number of distances we hypothesis that by modelling the feature as a continuous function of scale we can obtain information as to the shape of this function and hence improve its discriminatory power. This hypothesis is compared to the traditional method of selecting a given number of the best single distance measure. It is found on the limited data set available, that the classification accuracy can be improved by modelling the texture feature in this way