The Parallel Segmentation Algorithm Based on Pyramid Image for High Spatial Resolution Remote Sensing Image

Image segmentation is the foundation of the object-based and automatic interpretation of remote sensing images, but the high-resolution remote sensing image data is generally large, for this problem, the traditional approach is generally processing in sub-block, and then merge the results, but because of the complexity of the nature object, the merging result is not satisfied, and the segmentation algorithm is often more complex to calculate time-consuming, and it affect the image automatic interpretation of real-time. In this paper, we propose a parallel segmentation algorithm based on pyramid image, first of all, we create the pyramid image and segment it with the initial homogeneous regions were got, it divide the data according to the initial homogeneous regions and segment them from the top of pyramid image to the bottom with data parallelism, and it improve segmented efficiency, at the same time, it can avoid the problem of "merging line" when merging of the segmenting results in different image block. Experimental results show that the result of this algorithm is almost the same as the result of Mean Shift algorithm segmentation case; it says that this algorithm is correct and reliability, it also shows that this algorithm is efficiency by comparing the use of time between serial segmentation and parallel segmentation.