A STUDY OF FEATURE EXTRACTION USING DIVERGENCE ANALYSIS OF TEXTURE FEATURES W. Hallada, B. Ely and R. Boyd Computer Sciences Corporation Silver Spring, Maryland 20910, U.S.A

This paper presents an empirical study of texture analysis for feature extraction and classification of high spatial resolution remotely sensed imagery (10 meters) in terms of specific land cover types. Little is known as to which texture features are important for separating specific land covers with a per-pixel classifier. The principal method examined is the use of spatial gray tone dependence (SGTD). The SGTD method reduces the gray levels within a moving window into a two-dimensional spatial gray tone dependence matrix which can be interpreted as a probability matrix of gray tone pairs. Haralick et al (1973) used a number of information theory measures to extract texture features from these matrices, including angular second moment (inertia), correlation, entropy, homogeneity, and energy. The derivation of the SGTD matrix is a function of: 1) the number of gray tones in an image; 2) the angle along which the frequency of SGTD is calculated; 3) the size of the moving window; and 4) the distance between gray tone pairs. In this study, the first three parameters were varied and tested on a 10 meter resolution panchromatic image of Maryville, Tennessee using the five SGTD measures. A transformed divergence measure was used to determine the statistical separability between four land cover categories— forest, new residential, old residential, and industrial—for each variation in texture parameters.