Image-based Occluder Selection

In many cases complex scenes are composed of densely occluded objects, which means that only a small fraction of the scene is visible, while the remaining parts are occluded. This property can be exploited to accelerate rendering by simply removing most of the hidden parts before sending the polygons to the graphics hardware. In recent years, a lot of so called occlusion-culling methods have been proposed to speed up real-time rendering that way. A common prerequisite to many of these methods are fast occluder selection algorithms. This work focuses on a novel approach to determine nearly optimal occluder objects with respect to arbitrary scenes. Instead of selecting objects by evaluating simple metrics like size, distance, polygon count etc., the selection is based on the examination of an early low-resolution image of the scene. This image is created by means of a simple image-based rendering technique. In using an image-based approach, the speed of the rendering process is not bound by the objects’ polygonal complexity. Thus, the rendering time can be minimized, independent of the number of polygons that the scene’s objects are composed of. The accuracy of the resulting image is still too low for an exact determination of the visible objects of the final image, but sufficient for improved occluder selection. By including additional information, e. g. perpixel depth complexity and the objects’ projection area, the objects found to be visible are further restricted to a minimum occluder-set. The high accuracy of the presented occluder selection approach facilitates the use of run-time efficient one-pass occlusion-culling methods. Such methods do not put high demands on the graphics hardware, which makes them an ideal solution for low cost graphics systems.