Toward the Labeled Segmentation of Natural Images Using Rough-Set Rules

This article introduces an approach that integrates color and texture features for the segmentation of natural images. In order to deal with the vague or imprecise information that is typically shown in this kind of scenes, our method consists in a supervised classifier based on rules obtained using the rough-set theory. Such rough classifier yields a label per pixel using as inputs only three color and three textural features computed separately. These labels are used to carry out the image segmentation. When comparing quantitatively the results from this work with state-of-the-art algorithms, it has shown to be a competitive approach to the image segmentation task. Moreover, the labeling of each pixel offers advantages over other segmentation algorithms because the outcome is intuitive to humans in two senses. On one hand, the use of simple rules and few features facilitate the understanding of the segmentation process. On the other hand, the labels in the segmented outcomes provide insight into the image content.