Spectral matching using pixel cross-correlograms for the analysis of LANDSAT TM data

Abstract

A technical description of a new approach for classification of LANDSAT Thematic Mapper data is given using a spectral matching algorithm based on pixel cross-correlograms. A pixel cross-correlogram is derived by calculating the cross-correlation at different match positions between a reference spectrum (ie, a laboratory or pixel spectrum known to represent a material of interest) and a test spectrum (ie, a pixel spectrum) by shifting the reference spectrum over subsequent channel positions. The correlogram for a perfect match is parabolic in shape with a peak correlation of 1 at the central match number. Quantification of the differentiation from this shape may be used to measure the similarity of the materials tested. Image classification is achieved using the correlation coefficient at match position zero, the moment of skewness of the pixel correlogram and the significance of the pixel correlogram from a “3D” cross-correlogram data cube which contains stacked correlation images for the different match positions. Validation of the classified images is implemented through a root mean square error approach.