An edge preserving filtering process based on anisotropic diffusion equation is proposed before image segmentation. This preprocessing filtering method has better segmentation results compared to the image smoothed by Gaussian filter or the original image without smoothing by any linear low-pass filter. The advantages of this approach include: 1) like any other linear low-pass filters, the image smoothed by this filter before segmentation, which will lead to a multi-scale representation, will result in fewer segmented regions than that of the original image. 2) this nonlinear smooth selective low-pass filter is adaptive to the local structures and enables accurate boundary localization. We apply this filter for high resolution satellite remote sensing image before its segmentation and achieve better segmentation results.
展开▼