Geodesic interactive segmentation in the color monogenic signal framework

Guillaume Demarcq (1) , Hoel Le Capitaine (2) , Michel Berthier (3)

1 Explora Nova
2 LINA - Laboratoire d'Informatique de Nantes Atlantique
3 MIA - Mathématiques, Image et Applications - EA 3165

In this paper, we present an interactive algorithm for segmentation of color images. The user first draws some scribbles into regions that must be discriminated, and the segmentation is then automatically obtained. The segmentation is based on the computation of geodesic distances within color monogenic signal (CMS) fields. An important difference with state-of-the-art methods is that scribbles, which are often segments, are sample pixels picked up by the user. It results in a much more user-friendly segmentation process. Experimental results and comparisons with recent methods show the effectiveness of the approach.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Guillaume Demarcq, Hoel Le Capitaine, Michel Berthier. Geodesic interactive segmentation in the color monogenic signal framework. IEEE Int. Conf. on Image Processing, Sep 2012, Orlando, United States. pp.1573 - 1576. ⟨hal-00732832⟩