Abstract: Natural image segmentation occupies an important position in areas such as image processing and computer vision. The image segmentation algorithm based on clustering is an important method in unsupervised image segmentation algorithm. However, these methods have two main problems. First, features extraction is usuallly based on pixels. This makes the segmentation results do not fit the edges of images. Super pixels are introduced to preprocess images to overcome this problem. Second, it’s difficult to determine the number of segments. An energy difference based on mutual information is proposed to determine the number of segments automatically. Experimental results on standard database show that the proposed algorithm conquers two problems and can segment natural images well.

Key words: image processing, computer vision, super pixels