Texture Feature Extraction of Multiband Remote Sensing Image Based on Gray Level Co-occurrence Matrix

In this paper, we discuss the problem of how to select the weights of every band in multiband remote sensing image when we extract texture features based on gray level co-occurrence matrix. Two methods are proposed: the first one is to use the theorem of the mean method and the second one is to use the principle component analysis method. According to the technology of the texture characteristic, the origin multiband remote sensing image is divided to many unoverlappecl subregions, which have the same number of rows, columns, bands and pixels. Feature vectors in every band can be composed to a single one as the feature vector of the subregion by using the theorem of mean method and the principle component method. For classification of the multiband remote sensing image, we suppose all the data in the texture space obey Gauss finite mixture model distribution, use the expectation maximization algorithm to estimate the parameters and use Bayes decision to get the classification of the subregion. Finally the results will be compared and best result can be obtained.