SCALE-BASED ROBUST IMAGE SEGMENTATION

Abstract image. Natural clutter for example, tends to consist of a large number of equivalued scat-Image segmentation represents an essential step terers, and this in contrast to the man-made in the early stages of an Automatic Target one, mostly comprising a few prominent scat-Recognition system. We propose two robust ap- terers. It is precisely this type of statistical proaches fundamentally based on scale informa- characteristic that we are interested in captur-tion inherent to a given imagery. The first ap- ing and subsequently using as the basis for our proach is parametric in that the scale evolution classification of various terrain types in SAR. of an image is statistically captured by a model In addressing such a problem, we have two ba-which is in turn utilized to classify the pixels sic approaches, the first aims at characterizing in the image. The second, on the other hand, the statistics of the image evolution in scale and nonlinearly evolves the image along some spe- ultimately using them in the pixel classification; cific characteristic to unravel and delineate the the second attempts to diffuse "noncomplying" various comprising entities in it. observations via a nonlinear evolution to make it converge to specific desired domains of attrac