A new texture approach to discrimination of forest clearcut, canopy, and burned area using airborne C-band SAR

In this paper, we review some of ways in which texture classification has been handled in the image processing literature and introduce additional texture measures, namely, structure features. The equations that define a set of two measures of structure features are given. One of them relates to the structure component of the image. Another characterizes the structure complexity that occurs in the image. The impact of the variations of structure feature on classification accuracy is checked by a stepwise regression method. A comparison of performance is made between structure features and Haralick´s features, extracted from the graylevel co-occurrence. The results suggest that structure features have discriminatory power in forest cover classification. Compared with Haralick´s feature measures, structure feature measures minimize internal calculations within windows and are relatively simple. The outcome of structure feature measures need not be standardized. This gives a “unit-free” measure and “naturally occurring” groupings of objects. In combination with a classifier, structure feature measures are a significant extension over conventional texture feature measures