Local Window K_means Clustering and Merging for Color Image Segmentation

In this paper, we propose a faster and more efficient color image segmentation technique, which is called local window K_means (LWK_means), consisting of three modules: window presetting, local window clustering, windows merging. LWK_means divides the color image into many windows, and then parallelly processes each window using the proposed local window K_means clustering algorithm, which is adaptive and gives a more reliable initial clustering center instead of random initialization, in compressed HSI color space. And the final windows merging method is proposed in such a way as to automatically pull the independent windows together into an image with good spatial continuity. Experimental results demonstrate that the proposed technique is able to work better than the state-of-the-art color image segmentation algorithms with the least time consumption achieving a higher efficiency.