Texture image classification and segmentation using RANK-order clustering

Image analysis using texture as a spatial feature can be employed to segment regions of a complex scene or in the classification of surface materials. The relationship between most textural images and their description is mathematically intractable. In this paper the authors propose a new statistical measure, which is not based on a pre-defined formulation. Here, the local information in all directions around a pixel and its neighbourhood is represented in a 'directional RANK-strength' vector. The proposed method leads to texture classification and segmentation methods. Both algorithms have been tested on natural images with results in agreement with perceived ones.<
>