Texture Classification and Retrieval by Adaptive Mean Shift Clustering and Edge Images

In the state-of-the-art approaches a texture is characterized through textons. The main idea of this method is to build a texton vocabulary and then use it to build a texton histogram for each image. The histogram is used to measure a similarity between images. Since the textons are centers of clusters in a high dimensional space built from a training image set, we need some instrument for the feature space analysis. As a clustering algorithm the adaptive mean shift algorithm was chosen. In our paper we assume that textures are 3D materials. This means that under different viewpoints and photographic conditions 3D textures can change their appearance significantly and thus can have quite different histograms. In this paper we propose a method which uses edge images instead of original for constructing textons vocabulary and texton histogram. Insignificant details and noise could also be reduced. The performance based on original images and edge images are compared and results are presented.