A knowledge-based segmentation technology for remote sensing optical images

In this paper, we propose a whole scheme of remote sensing image segmentation process, from fast detection to accurate edge location. As we know, more structure information is acquired in high resolution remote sensing images. However, traditional image processing algorithms will produce meaningless results without priori knowledge. We aim at solving the problem in which regions may be distinguishable in intensity but belong to the same target by the ground truth. This is done by multi-threshold segmentation. What's more, In order to get a more regular shape, we use random field model to introduce spatial constraint at a small scale, and active contour model to smooth the whole edge at a larger scale. Simulation results demonstrate the effectiveness of our method in extracting ships from the satellite images. This paper also introduces the potential of integrating the image segmentation and subsequent image analysis tasks.