Smoothing with adaptive rank order filters for early processing in image segmentation

This paper describes two non-linear smoothing filters based on adaptive rank order filtering. the ideal output of these filters is a grey value suitable for characterizing the grey value distribution associated with the local texture, i.e. suitable for use as a region label for image segmentation purposes. the problem is to select as output a grey value that is stable with respect to the fluctuations of the local grey value distribution inside homogeneous regions. the first proposed method aims at the estimation of the mode of the local distribution. the second selects the output rank in dependence on the local average; it is simpler and combines the edge-sharpening advantages of ‘comparison and selection filters’ with a better mode estimation, in comparison with the median filter, in the presence of asymmetrical local grey value distributions.