Method for tree-structured hierarchical occlusion in image generators

Abstract

A method for determining which objects are visible in each pixel of a digitally-generated image is described. The method uses comparison of distances from the viewpoint (Z-sorting) for occlusion. The number of Z comparisons required is minimized by utilizing a hierarchy of regions in the screen area, and thereby, resolving occlusion for areas larger than pixel areas where possible. Use of the area hierarchy increases the efficiency of the occlusion processing. In a preferred implementation, the screen is divided into equal-sized horizontal regions each corresponding to typically four scanlines. Polygons for each image are subdivided into pieces that correspond to the preset hierarchy of regions, and occlusion processing occurs independently in parallel for two or more regions. Each region is subdivided further into smaller regions, each level of the hierarchy having regions half the size of a higher level region. The minimum and maximum Z values of a polygon's contribution to a region are compared to previously determined values for a hierarchy of subregions. Occlusion is accomplished at the highest level of the hierarchy at which portions of the incoming polygon contribution either occlude or are occluded by previous polygons. Provisions are made for transparent and interpenetrating polygons.