A Watershed Algorithm Combining Spectral and Texture Information for High Resolution Remote Sensing Image Segmentation

High resolution remote sensing image segmentation methods that consider only the spectral information in the region growing process often lead to over segmentation and low boundary precision. To overcome that, a watershed transform algorithm which combines spectral information and texture information is proposed. At first, the spectral intensity gradient and the texture gradient have to be extracted from the input image. For that purpose, a new bilateral filtering model is introduced. This edge preserving algorithm can remove noise of images. Meanwhile, it can also remove texture from images by using a local smoothing scale parameter. By adapting this filtering algorithm on the original image and the Gabor texture feature images, the spectral information and texture information are extracted separately. Then with edge detection algorithm, the spectral intensity gradient and texture gradient are obtained. Finally a gradient fusion strategy by morphological dilation and watershed transform are performed in succession. Experiments are carried out on three high resolution color remote sensing images. Compared with JSEG and multi-resolution segmentation methods, the proposed method has a higher boundary precision and can reduce the over segmentation and under segmentation effects.