Segmenting omnidirectional images

Le résumé fourni par la source

We define in this paper an image segmentation model for omnidirectional images, i.e. for images containing information of a scene in all directions simultaneously. The segmentation model is based on the extension of active contours/snakes to omnidirectional images. We start by considering omnidirectional images generated by spherical and hyperbolic mirrors, whose conformal equivalence makes it easy to alter the initial theory. We then describe a broader technique that is able to generalize the evolution of active contours on a parabolic mirror and a larger class of parametric manifolds as well.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.