Segmentation is the basis of object-oriented image analysis. For many years, due to the increasing require for object-oriented image analysis, procedures for image segmentation have been a main research focus in this area. The aim of segmentation is to extract the interesting region from remote sensing image, so the region-based approach is the best way for image segmentation. The region growing method generates many meaningful objects through merging the spectral-similar neighboring pixels. Meanwhile landscape spatial heterogeneity requires multi-scale analysis with remote sensing information extraction. Each pattern or process has its inherent feature in different scales. To ensure high precision surface information, the remote sensing application model building on one scale image need to be modified if it is used on the other scale. The combination of image segmentation and multi-scale analysis becomes a new trend in remote sensing application. Based on the scale affect and minimum-heterogeneity rule, this paper presents the necessity and possibility of multi-scale affects analysis as well as the principle and practice of the region growing image segmentation. There are two sites to test the multi-scale image segmentation process. The results show the image objects richness of geometry and semantic information. Therefore this approach offers an optical solution for the object-oriented and multi-scale image analysis.
Progress in Geography