This paper presents a method for image segmentation by clustering superpixels based on histogram dissimilarity. A superpixel is a tiny image fragment which contains the edge or boundary information of sub-regions in the image. Each sub-region consists of multiple superpixels which have a similar texture. In the proposed method, K-means clustering algorithm based on the color histogram dissimilarity is used to merge superpixels to a sub-region. A dissimilarity between two color histograms calculated by the divergence measures. Experimental results show that the proposed segmentation method extract sub-regions reasonably from images.