Selective Rendering using Task-Importance Maps

摘要

The perception of a virtual environment depends on the user and the task the user is currently performing in that environment, see figure 1. This poster presents a novel approach for creating and using task maps in selective rendering to control image detail. This a priori knowledge of task-level saliency may be used to optimize the computation of animations, saving significant rendering time without altering the user's perception of the scene. Perception-guided rendering algorithms exploit models of the Human Visual System (HVS) to significantly reduce computational time when rendering realistic images, without compromising the perceived visual integrity. There are two general visual attention processes, labeled bottom-up and top-down [James 1890], which determine where humans locate their visual attention. In bottom-up processing the visual stimulus captures attention automatically without volitional control. In contrast, the top-down process is under voluntary control, and is relevant to the observer's goal when studying a scene. The saliency map (SM) [Itti et al. 1998] indicates the most salient, or visually important, regions in a given image based on the low-level factors color, intensity and orientation in a purely bottom-up manner. Task maps (TM) on the other hand specify what parts of a scene are perceptually important for performing the task [Haber et al. 2001; Cater et al. 2003].