Temporal Anti-Aliasing and Temporal Supersampling in Three-Dimensional Computer Generated Dynamic Worlds

Abstract: This master thesis investigates and evaluates how a temporal component can help anti-aliasing with reduction of general spatial aliasing, preservation of thin geometry and how to get temporal stability in dynamic computer generated worlds. Of spatial aliasing, geometric aliasing is in focus but shading aliasing will also be discussed. Two temporal approaches are proposed. One of the methods utilizes the previous frame while the other method uses four previous frames. In order to do this an efficient way of re-projecting pixels are needed so this thesis deals with that problem and its consequences as well. Further, the results show that the way of taking and accumulating samples in these proposed methods show improvements that would not have been affordable without the temporal component for real-time applications. Thin geometry is preserved up to a degree but the proposed methods do not solve this problem for the general case. The temporal methods' image quality are evaluated against conventional anti-aliasing methods subjectively, by a survey, and objectively, by a numerical method not found elsewhere in anti-aliasing reports. Performance and memory consumption are also evaluated. The evaluation suggests that a temporal component for anti-aliasing can play an important role in increasing image quality and temporal stability without having a substantial negative impact of the performance with less memory consumed.