Methods Of The Image Segmentation Based On The Spatial Autocorrelation

Image segmentation is a hot issue of digital image processing. Now, in the field of image segmentation, using specific professional knowledge to complete segmentation is one of its characteristics. Results of research show that spatial autocorrelation can also be applied in image segmentation area. In the first place, this dissertation outlines the current status of image segmentation, the development process of spatial autocorrelation analysis and several common spatial autocorrelation indexes. Based on random gray image, this dissertation has designed a series of different experiments, compared the function and characteristics of various types of spatial autocorrelation indexes, and completed the boundary identification between single nuclear assembly and multi-core clustered which is based on the random image. Studies show that global Moran's I is better than global Getis G from a global point of view, but from a local point of view local Getis Gi is better than local Moran's Ii. This dissertation used spatial autocorrelation indexes which were selected by basic experiment, combined with iterative threshold selection method and derivative operator, has completed boundary identification of trunk road with Quick Bird high resolution image of Fu Zhou. Compared with original image, Experimental results has a certain accuracy. Therefore, image edge recognition can be done based on the results of spatial autocorrelation analysis in practical applications.