A practical method is proposed for extracting and classifying regions in remotely sensed multi-spectral images. First, an entropy index based on the gray level histogram of local sub-area is shown to be effective for extracting regions which have homogeneous gray levels in the image. It is demonstrated in the next step that the maximum likelihood method using spectral characteristics of the extracted regions is capable of classifying regions in the image. This multi-stage unsupervised method is applied to the actual environmental land use classification.
Read full abstract