Texture segmentation using a dynamic iterative self-organizing clustering algorithm

One of the most critical low-level layers in computer vision systems is the Segmentation Layer. In this paper, a dynamic iterative clustering algorithm is proposed based on the ISODATA algorithm. The main drawbacks of the ISODATA algorithm are its need for an initial estimate of the number of clusters, K-0, and the dependency of its results on that number and on the selection of the initial cluster seeds. This paper proposes some modifications to eliminate the first drawback by estimating a suitable initial number of clusters and providing the dynamics necessary to change that estimate during the clustering process. To reduce the effect of the initial seeds, an informed method for selecting them is used. The algorithm has been used in a Texture Segmentation Layer using a set of textural features selected to segment some sample images from a specific class of scenes. Promising results are reported and investigated in this paper.