Example-Based Segmentation of Satellite Images Using Detection of Region Boundary

Abstract

This paper proposes a method of example based segmentation using detection of region boundary to segment the regions such as a city, a mountain, and the river in satellite images. These regions are classified as texture regions. Example based Segmentation is performed with characteristic models based on texture analysis by using local fractal dimension . Unfortunately, this method segments the region in the boundary with a low accuracy. To solve this problem, it is necessary to detect the region boundary. The region boundary is detected by using Canny operation and mathematical morphology. The output image obtained by integrating the example based segmentation image with the image in which the region boundary is detected improves accuracy of the segmentation in the region boundary. We show the effectiveness of the proposed method in comparison with the correct image we make with a manual from the input image.