A new process for the segmentation of high resolution remote sensing imagery

The watershed transformation is a well-known powerful tool for automated image segmentation. However, it is often computationally expensive and can produce over-segmentation in situations of high gradient noise, quantity error and detailed texture. Here, a new method has been designed to overcome these inherent drawbacks. After pre-processing the imagery using a nonlinear filter in order to filter the noise, an optimized watershed transformation is applied to provide an initial segmentation result. Then, a multi-scale, multi-characteristic merging algorithm is used to refine the segmentation. Preliminary results show promise in term of both segmentation quality and computational efficiency.