Feature distributions for unsupervised color texture segmentation

This paper proposes an unsupervised color texture segmentation method that conjoins the feature distributions of color features and local texture patterns to derive the homogeneity measure for partitioning the regions of color image during the process of segmentation. Two feature distributions are used in this paper to distinguish different regions of color textures, namely a fuzzy color histogram and the texture feature number histogram. The former is related to the distribution of color features, while the latter is related to the distribution of local texture patterns in a texture region. A region-based coarse-to-fine algorithm based on the proposed homogeneity measure is employed for coarsely segmenting the regions of color image, and then a pixel-wise classification scheme for improving localization of region boundaries. The feasibility and effectiveness of the proposed method is evaluated with the various types of test images that include the collages of real texture and the natural scenes in the experiments.