抄録
Image segmentation is not only very important technique, but also challenging issue in Computer Vision. The segmentation of an image and grouping of the segmented regions as an object are very hard task for a computer, but they play a great role in human visual perception. There are, therefore, many researches in the field of color image segmentation. It is very difficult task to make a closed boundary for a region from obtained edges using edge detectors. Active contour model effectively extracts a boundary of an object, but initial boundary affects the final results. Moreover, statistical models have optimization difficulties. In this paper, a region growing approach which employs several internal features in a region and difference between two adjacent regions is proposed. Small regions occur on irregular boundaries or texture regions by using the conventional region growing segmentation approaches. Our approach reduce the small regions by using size, color distributions, edge magnitude and average color of regions. Moreover, the proposed method extracts segments suitable for human visual perception. The paper shows some examples on both synthetic and real images.
