An Image Segmentation Algoithm Based On Edge And Heterogeneity Of Objects

With the emergence and development of commercial high-resolution satellite, represenratives of which are QuickBird, IKONOS, WorldView GeoEye, etc, the development of space technology and spatial information industry was promoted rapidly, but it also brings great challenges. In order to make the high spatial resolution remote sensing images transform data into information, and then it could serves for various fields, thematic information data that all fields mentioned required must be extracted. Information extraction and target recognition were proposed under this requirement and then become hot points in high spatial resolution remote sensing image data processing.Traditional pixel-based information extraction technology to extract has significant limitations, and object-oriented image analysis methods in the elimination of salt and pepper noise and processing spatial relations have a great advantage. This method not only improved the accuracy of the extraction, and greatly shortened the time of information extraction, and it had more advantages than traditional classification algorithms no matter in extraction accuracy or in time efficiency. However, object-oriented image analysis method was based on the object as a unit, and the object was obtained by image segmentation, so the following information extraction accuracy would be directly affected by segmentation results. Now, there was not a uniform and good universal one in image segmentation methods. In view of this, in this paper mathematical morphology, edge detection and object heterogeneity were introduced into the process of high spatial resolution remote sensing image segmentation.In the paper, we combined the color, shape and overall heterogeneity of high spatial resolution remote sensing image with mathematical morphology, neighborhood processing and edge detection, etc, and then we integrated mathematical morphology, edge detection and object heterogeneous, etc. Some work like theoretical discussion, algorithm design and experiment analysis, etc had been done based these. The contents are as follows:(1) A canny segmentation algorithm improved by mathematical morphologytic was proposed. Firstly mathematical morphology was used to eliminate nosie of high spatial resolution remote sensing image in the algorithm, and then the Canny was used for edge detection. The method could retain more edge information and improve segmentation accuracy;(2) A multiresolution segmentation algorithm improved by mathematical morphologytic was proposed. Smash and over-segmentaion of image object which was segmentated by traditional multireolution segmentation were serious.However, above problems could be solved after using mathematical morphology to optimize the background information and noise of high spatial resolution remote sensing image.(3) An image segmentation algorithm based on edge and heterogeneity of objects was proposed. The edge information was combined with multiresolution segmentation in this method. The image objects were got by using edge information. Image segmentation accuracy was improved greatly by this method.