A unified information-theoretic framework for viewpoint selection and mesh saliency

Abstract

Viewpoint selection is an emerging area in computer graphics with applications in fields such as scene exploration, image-based modeling, and volume visualization. In particular, best view selection algorithms are used to obtain the minimum number of views (or images) in order to understand or model an object or scene better. In this article, we present a unified framework for viewpoint selection and mesh saliency based on the definition of an information channel between a set of viewpoints (input) and the set of polygons of an object (output). The mutual information of this channel is shown to be a powerful tool to deal with viewpoint selection, viewpoint stability, object exploration and viewpoint-based saliency. In addition, viewpoint mutual information is extended using saliency as an importance factor, showing how perceptual criteria can be incorporated to our method. Although we use a sphere of viewpoints around an object, our framework is also valid for any set of viewpoints in a closed scene. A number of experiments demonstrate the robustness of our approach and the good behavior of the proposed measures. ACM Reference Format:Feixas, M., Sbert, M., and González, F. 2009. A unified information-theoretic framework for viewpoint selection and mesh saliency.

Viewpoint information channel for illustrative volume rendering

“…In [26], it has been shown that one of the main properties of VMI is its robustness to deal with any type of discretisation or resolution of the volume data set. The same behavior can be observed for polygonal data [6]. From the viewpoint information channel, the viewpoint entropy (VE) of viewpoint v is defined by…”

“…Viola et al [26] proposed a viewpoint information channel and used the viewpoint mutual information to automatically determine the most expressive view on a selected focus. A unified information-theoretic framework for viewpoint selection, ambient occlusion, and mesh saliency for polygonal models has been presented in [6,7]. Next we review the definitions of viewpoint information channel, viewpoint mutual information, and viewpoint entropy in volume visualization.…”

“…The term 'balance' is used here to express that the visibility distribution p(Z|v) of v is similar to p(Z). This similarity is expressed by the Kullback-Leibler distance between p(Z|v) and p(Z) (see [6,26]). This distance is zero when p(Z|v) = p(Z).…”

“…This method has been previously used to select the most informative views for polygonal models [6]. The information projection over a viewpoint v can be done weighting the VOMI of voxel z by the transition probability p(v|z) and summing over all voxels.…”

Visual saliency guided textured model simplification

“…It can improve 3D model examination by evaluating view-points through Jensen-Shannon divergence [7] to reveal important model features. It can also identify and emphasise important parts of a volumetric model [18], or can modify the normal vector of each mesh vertex according to the variation in its luminance in order to enhance the visual appearance of important mesh features [30].…”