Graph Based Multispectral High Resolution Image Segmentation

An improved multispectral high resolution object-based image segmentation method based on graph is presented in this paper. We use Minimum Span Tree optimal theory to realize object based high resolution image segmentation. First, we improved the calculation of edge-weight by introducing the band-weight and Normalized Difference Vegetation Index (NDVI). Second, we proposed an edge based auto threshold select method which can realize multi-scale image segmentation by changing the scale parameters. The QuickBird multispectral images were used to do the image segmentation experiment. The experiment result proved that this method can obtain high quality segmentation, which can remain the detail of object, and this algorithm can be efficiently used in remote sensing images. It is also shown that the improved segmentation scale set method is easier to control for the user and for any kind of image.