The sensitivity of texture measures for correlated random clutter

In previous papers a variety of natural clutter textures have been demonstrated to be consistent with a correlated noise process having K-distributed intensity statistics. Fundamental to such a noise model is the expectation that all the correlation properties of such a process can be completely described in terms of its autocorrelation function (ACF). The ACF thus provides the basis for discriminating between such textures. As demonstrated previously, expressions for the intensity ACF can be derived analytically from the surface ACF. Thus a solution to the inverse problem of characterising the cross section fluctuations is to deduce the values of the parameters which describe the noise process by fitting the intensity ACF. The effectiveness of this fitting process in turn depends on the choice of texture discriminant. The author analyses the performance of two texture discriminant measures based on the ACF, namely the mean-square ACF difference and the normalised ACF cross-correlation measure.