Homogeneity of Pixel Neighborhoods in Gray Level Images Investigated by the Grade Correspondence Analysis

Abstract

The paper presents some interesting properties of Grade Correspondence Analysis (GCA) used to order the pixels of gray level images described by their gray level (gl), gradient module (gm) and a family of variables dealing with pixel neighborhoods. The gm stripcharts for pixels ordered according to GCA are used to separate and then order subsets of pixels with identical descriptions of their gm neighborhoods. The spatial structure of clusters formed from subsets of adjacent clusters is then restored and used to discover compact areas of pixels with similar gm neighborhoods. Such a decomposition is performed for a MR image of a fragment of human brain. Two similar homogeneous compact areas discovered in this image suggest that the proposed method might be useful in medical classification and identification.