Category classification of multispectral image data using spatial information in the small image region

A new category classification method of high spatial resolution multispectral image data using the spectral and spatial information of the small image region is proposed. The gray level mean and standard deviation of pixels in a 3×3 image region were defined as the spectral and spatial information of the center pixel, respectively. And these were combined in the feature space with the second order moment about the origin. To compare the classification performances of the proposed method with the conventional methods, category classifications were performed with an actual image data of Landsat-5 TM. There was a substantial improvement in overall classification accuracies with the proposed method