Satellite Cloud Image Segmentation Based on the Improved Normalized Cuts Model

We propose a novel approach for satellite cloud image segmentation based on the improved Normalized Cuts Model. We extracted three important features from the multi-channel grayscale information and the texture features of satellite image, by the statistical analyses of the surface observation. Having set up the weight matrix by those features, we use the spectral graph theoretic framework of normalized cuts to find partitions of the satellite image, and estimate the cloud type of every partition. Experimental results on a satellite cloud image are shown.