A multi-scale and morphological gradient preserving contrast
Résumé
This document outlines an algorithm which extends and enhances the regularised morphological gradient. The latter has little noise, is multi-scale and therefore has been extensively used, for instance in the extraction of road markers from particularly challenging images. However, the intensity values taken by the regularised gradient are usually not representative of the perceived contrast between objects boundaries. This limitation may be problematic to some applications, because hierarchical segmentations obtained by means of waterfalls or synchronous flooding mechanisms on such a gradient would not match the human perception, since the hierarchy depends on the gradient first overflow zones. The proposed enhancement of the regularised gradient preserves the thinness and multi-scale properties of the latter whilst taking account of the actual contrast across different image scales.
Origine | Fichiers produits par l'(les) auteur(s) |
---|