A Survey of Viewpoint Selection Methods for Polygonal Models

Abstract

Viewpoint selection has been an emerging area in computer graphics for some years, and it is now getting maturity with applications in fields such as scene navigation, scientific visualization, object recognition, mesh simplification, and camera placement. In this survey, we review and compare twenty-two measures to select good views of a polygonal 3D model, classify them using an extension of the categories defined by Secord et al., and evaluate them against the Dutagaci et al. benchmark. Eleven of these measures have not been reviewed in previous surveys. Three out of the five short-listed best viewpoint measures are directly related to information. We also present in which fields the different viewpoint measures have been applied. Finally, we provide a publicly available framework where all the viewpoint selection measures are implemented and can be compared against each other.

3D Viewpoint Estimation Based on Aesthetics

“…This benchmark provides a comprehensive evaluation method based on ground-truth data and the 21 most advanced methods were compared. Bonaventura [55] provided comparative experimental results for 21 state-of-the-art methods and compared them with the best viewpoint selection results of the 26 subjects in the Dutagaci benchmark [54]. Bonaventura [55] provided an implementation framework and source code for 21 of the most advanced methods.…”

Exploiting user interactivity in quality assessment of point cloud imaging

“…In our analysis, to address how many perspectives are sufficient and what is the impact of enabling additional views, a model is captured by K regularlyspaced viewpoints with the following camera layouts: (a) a single point that captures the frontal view of the content (i.e., K = 1), to examine whether a single image corresponding to the initial view of the model that was displayed to the subjects provides a good approximation of its visual quality; (b) the vertices of a surrounding octahedron (i.e., K = 6), which is idential to the setup of [7]; and (c) points lying on a surrounding geodesic sphere with coordinates determined by iterative subdivisions of a regular icosahedron up to 2 levels (i.e., K = 12, 42, 162). The latter is a commonly used arrangement in studies for view selection [11], [12], that provides a consistent approach to approximate uniformly distributed samples that are lying on the surface of a sphere. By iteratively subdividing the regular icosahedron, gradual granularity with progressive integration of new viewpoints on the previous set is achieved; this is important in order to identify whether additional views can improve the prediction of subjective visual quality.…”

“…To the best of our knowledge, weighted views have been considered only in [11] for objective evaluation of 3D meshes. The importance weights were obtained based on a surface visibility algorithm, typically used for viewpoint preference selection [12]. In our analysis, provided an interactive subjective evaluation scenario, we make the hypothesis that the importance of a view is related to the duration of inspection from participants during subjective assessment; thus, the projected images are weighted accordingly.…”

Canonical 3D object orientation for interactive light-field visualization

“…Bonaventura et al recently presented a survey of the computational measures that focus on the "goodness for recognition". 3 In their survey, they tested a common framework for the most basic measures introduced to select the best views for polygonal models.…”

“…1 Knowledge of the canonical views or preferred orientations of 3D models helps several applications, such as automatic camera replacement, 3D scene generation, 2 scene navigation, scientific visualization, object recognition, mesh simplification, surgery planning, view-based 3D object recognition etc. 3 Several studies 4-10 have shown that orientation preference or view-canonicity is dictated by multiple factors such as: recognition, familiarity, functionality and aesthetic criteria. To seek preferred views, the researchers have adopted different methods, including ranking photographs taken from several viewpoints, imagining a viewpoint of an object based on its name followed by photographing the object to obtain same view, and orienting 3D models displayed on a 2D screen, using a controller.…”