Perceived Realism in Computer Graphics : Evaluating Virtual Snow Accumulation with Two-Alternative Forced Choice

Sammanfattning

Snow simulation has a nearly 30-year history in computer graphics. As computer graphics have become more ubiquitous, methods of snow generation have improved, creating implementations that can automatically generate lifelike snow in scenes. These are especially useful for film, animation, and game development. Rarely are these implementations formally evaluated on how good they are at generating visually appealing snow. In this paper, a voxel- and physics-based implementation for generating snow over 3D scenes is presented. Snow covers created by the implementation are then evaluated by participants using a two-alternative forced choice (2AFC) tool. The participants compare renders created with different settings on three different models (sphere, tree, house). The evaluation finds that different settings influence the perceived realism differently. Which settings and how strongly they are preferred differs with the model chosen, with more visually chaotic models being harder to spot differences in and leading to less strong preferences.