Improved watershed segmentation algorithm for high resolution remote sensing images using texture

Abstract

Image segmentation, the combination of image elements based on homogeneity of the same segments and the heterogeneity to the neighboring regions, is particularly important in image processing. The difficulty in the identification of such property leads to the complex of the image segmentation. Among the segmentation approaches, the watershed algorithm has been extensively employed. However, the over-segmentation and under-segmentation have become the key problems for the conventional algorithm. To improve this method, texture of high resolution remotely sensed data are employed. As a case study, the QuickBird image has been segmented by using the improved approach. In the segmentation process, texture is regarded as a separate 'band'. Finally, the effectiveness of the improved watershed segmentation approach for high-resolution imagery has been analyzed. And the approach to solve the problems such as over-segmentation and under-segmentation has been proposed.