Image segmentation by mapping the local histograms into a region-label space

The problem addressed in this contribution is that of segmenting a natural image into homogeneous regions by `tagging´ each pixel with a proper region label, which is provided by a symbolic grey value. The author proposes an alternative approach, which is advantageous in all the cases in which, for characterizing a region, the pixel geometry is much less substantial than the complete shape of the local grey value histogram. Instead of dealing with the set of spatially ordered grey values, this approach utilizes the set of the rank-ordered local grey values (i.e. the local histogram), renouncing to whatsoever feature extraction