Abstract
We propose a method for segmenting a color image into object and background regions by a computer. The task of the method is extracting unknown objects which consist of some color regions without any help of users. The image is initially segmented into small regions by k-means clustering. Next, small regions are repeatedly merged by five merging rules. The merging rules are based on information about positional relationship or shape of small regions, and "the degree of focus" showing whether the region was in focus or not when the picture was taken. We conducted experiments with actual images to compare to the conventional method which has the similar problem setting. The results show that the proposed method is better in terms of the precision of boundaries.
