A figure extraction method based on the color and texture contrasts of regions

Detecting interesting regions from pictures has become important in order to reduce the computational complexity associated with such time-consuming processes as object recognition. In this paper we assume that figure-ground segmentation of a picture is necessary as the first step towards evaluating the attractiveness of regions. In vision research, it has been found that the V4 cortex in the visual system contributes to figure-ground selectivity. By utilizing this finding, we propose a method for extracting figure regions from a picture based on the color and texture contrasts of regions. This method discriminates regions previously segmented from a picture into figure regions and ground regions by a discrimination function defined with contrast parameters. The contrast parameters include the color difference and the texture difference between a region and its surroundings, as well as the focus of the region. A discrimination function is constructed based on the characteristics of the extracted figures and ground obtained in our experiments. In other experiments on evaluating the accuracy of the proposed method, figure regions selected by human subjects are extracted by the method at the rate of 80% accuracy.