Spectral and textural classification of single and multiple band digital images

Abstract Single and multiple band images are classified using supervised algorithms. Two programs, MXTEXT and MXMULT, are presented that use minimum-distance-to-mean or Bayesian, maximum likelihood algorithms for spectral classification (pattern recognition), further allowing classification of image texture based on the local variogram surrounding each image pixel. Classification of texture can be performed independently of the classification of spectral information, or a combined spectral/textural classification can be performed. Combining the classification of texture with that of spectral information is shown to be of particular value for single band radar images. Improved classification accuracy is demonstrated also for multiple band images when variograms and cross-variograms are used to classify texture and spectral information.