Automatic texture segmentation based on k-means clustering and efficient calculation of co-occurrence features

This work presents a method for automatic texture segmentation based on the k-means clustering technique and co-occurrence texture features. A set of eight features were extracted from the grey-level co-occurrence information. Two of them are proposed here to improve segmentation results for magnetic resonance (MR) images. All the features are used to segment image regions based on textural homogeneity of its areas. As the process of calculating the grey-level co-occurrence matrix (GLCM) demands increased computational resources, we propose a new methodology based on an indexed list (IL) for fast element access. This novel approach highly optimizes the algorithm called grey level co-occurrence indexed list (GLCIL). Moreover, we compare the efficiency of the proposed method against two others, namely the traditional GLCM and the grey level co-occurrence linked list (GLCLL), which was suggested as a faster alternative to GLCM. The technique proposed here is the most efficient in terms of computational time, when compared to the other two. Additionally, examples on real time segmentation are presented to illustrate the appropriateness and robustness of this new method.