Detection of noisy regions with known frequency distributions

Image segmentation of images generated by (side looking) radar is difficult because of the low signal to noise ratio caused by the reflection of coherent radiation. Classical methods are based on the assumption that the regions have constant properties in some sense, and filters have been designed for testing this homogeneity hypothesis. Two weak points of the classical approach are : firstly the operator window size is fixed and secondly the noise is not modelled explicitly. The authors propose a method which is based on the conditional signal|object probability density distribution and which can be used to detect regions with different frequency distribution effectively.<<ETX>>