Knowledge Based High Resolution Remote Sensing Image Segmentation

This paper proposes an effective approach to extract linear object in high-resolution remote sensing image. The approach integrates the knowledge about the linear object to implement the watershed algorithm and guide the region merging and finally extract the linear object. First, the Kalman filter algorithm is used to detect the straight line in the image, the center point of parallel line pairs and the minimum with dynamics larger than predefined threshold are utilized as marker point to modify the morphological gradient of the input image by geodesic reconstruction, the modified gradient image is then segmented by the watershed transform. The initial segmentation result is input to region merging process. This process applies the region adjacency graph (RAG) representation of the segmented regions and knowledge about the road to execute the region merging, which significantly reduce the merging time. The proposed scheme was tested on remote sensing images of 2 m resolution, and the results show that the extraction of road is quite promising.