Edge Filtering Using Orientation Entropy

The paper introduces the concept of orientation entropy for quantifying the propensity of a pixel belonging to the edge. Intensity variation at a pixel is measured as an integral of a logarithmic function over all possible direction which represents the heterogeneity of intensity variation in the neighborhood of that point. This is unlike traditional edge filters which uses divergence operator considering the maximum intensity variation among all possible directions. The proposed entropic filter is found to provide visually superior edges for a number of benchmark images considered in our experiments