EFFECT OF NOISE ON ORDER-PARAMETER ESTIMATION FOR K-DISTRIBUTED CLUTTER

The paper addresses the characterisation of high resolution radar imag e textures in the presence of additive noise, which is inevitably pres ent in the system. Two possible goals are analysed. In the first, the authors consider absolute texture description and identify the extent to which noise degrades performance by introducing a bias. The second is concerned only with segmenting the texture into locally different r egions and discusses the effect of the noise on the sensitivity of the measure to texture changes, described in terms of relative variance. Initially, the authors demonstrate that estimates of the mean, normali sed log and contrast of the intensity approximate a sufficient statist ic for K-distributed clutter. They then compare the performance of a v ariety of texture measures in terms of the bias in the estimated order parameter for absolute classification and the relative variance for t exture segmentation. A normalised log measure is shown to have the bes t sensitivity overall. However, with additive noise an amplitude contr ast measure yields a much smaller classification error with only sligh tly reduced sensitivity.