Research On Element Combination Image Segmentation Method Based On Minimum Criterion Of Heterogeneity

The mushroom growth of high spatial resolution remote sensing technology not only takesopportunity but also challenge. The limitation of traditional ways of image analysis methodwhich based on pixel level is becoming more and more obvious. Object-oriented imageanalysis method，compared with the method which based on pixel level，do not only usespectral information but make full use of spatial information of RS images. This method winssuperiority in removing salt and pepper noise and handling spatial relationship，and now hasbecame the development tendency of remote sensing analysis technology.Image segmentation is a key step of object-oriented image analysis technology. In thecurrent study of image segmentation, there are several problems. Such as, the segmentationsare not aiming at high-resolution remote sensing image, the selection criteria of optimalsegmentation scale has not been built. This paper starts with multi-scale features of highspatial resolution remote sensing images，then did much jobs on theoretical discussion，algorithm design and experimental analysis. The main details as follows:(1)On the multi-scale hierarchy analysis of remote sensing image basis. This paper dwells onelement combination image segmentation method which based on minimumcriterion of heterogeneity. Then make it realized by c++programming method.(2)Using my own program as test platform, this paper analyses several important parameterswhich relate to the image segmentation method. And the conclusion is, the segmentationresult becomes finely when spectral factor rises. When shape factor rises, object outlinebecome regular, but this way may loss spectral information. Scale factor rises the scale ofobjects become larger. The optimal segmentation scale can be determined by means of meanvariance analysis.(3)Compared the segmentation method of this paper with another three mature segmentationmethods by the way of experiments. This paper finds element combination imagesegmentation method, which based on minimum criterion of heterogeneity, fitsbetter for the analysis of high-resolution remote sensing image.