Measuring similarity between color image regions

This paper presents a methodology for measuring the similarity between two color image regions. An isolated object or a segmented region is described with the Color Angular Radial Transform (CART) introduced in this study. This new region-based descriptor uses both the shape information in the luminance component and the spatial distribution of the dominant colors in the region. The similarity between two regions results from the weighted sum of the shape and color features. Results obtained on MPEG-7 data set are presented and discussed.