A Cost-Effective Supersampling for Full Scene AntiAliasing

Abstract

We present a graphics hardware system to implement supersampling cost-effectively. Supersampling is the well-known technique to produce high quality images. However, rendering the scene at a higher resolution requires a large amount of memory size and memory bandwidth. Such costs can be alleviated by grouping subpixels into a fragment with a coverage mask which indicates which part of the pixel is covered. However, this may cause color distortion when several objects either overlap or intersect with each other within a pixel. In order to minimize such errors, we introduce an extra buffer, called the RuF(Recently used Fragment)-buffer, for storing the footprint of a fragment most recently used in the color manipulation. In our experiments, the proposed system can produce high quality images as good as supersampling with a smaller amount of memory size and memory bandwidth, compared with the conventional supersampling.