A multi-scale and morphological gradient preserving contrast
Abstract
This document outlines an algorithm which extends and enhances the regularised gradient. The regularised gradient is known to be a thin gradient. It has little noise, is multi-scale and has been extensively used, for instance, for the extraction of road markers from particularly challenging image sets. However, the intensity values taken by the regularised gradient are usually not representative of the perceived contrast across object boundaries. This limitation may prevent hierarchical segmentations based on the waterfalls or synchronous floodings from being meaningful with respect to the actual human perception of saliency. The proposed enhancement of the regularised gradient preserves the thinness and multi-scale properties of the latter whilst taking the actual contrast across different image scales into account.
Origin : Files produced by the author(s)
Loading...