A Fully Automatic Colorimetric Saliency Detection Approach for 3D Meshes

Abstract

With the democratization of acquisition systems, 3D meshes have established themselves as the most optimal modality for representing, analyzing, transmitting, editing and printing 3D contents. All these processes have one thing in common: the human observers who visualize the 3D content downstream. To optimize these different processes, it would be judicious and relevant to take advantage of a characteristic present in all human visual systems, namely the sensitivity to visual saliency. In the literature, very little work has been carried out on predicting visual saliency on colored meshes, as opposed to uncolored meshes. Yet the colorimetric aspect of visual content (whether 2D or 3D, static or moving) plays a key role in directing the human visual attention. In this paper, we propose a new approach, requiring no intervention, parameterization or learning, for the detection of visual saliency of 3D colored meshes. The results demonstrate the relevance and the superiority of the proposed approach.