Rasterization by Multiresolution Integration

Abstract

Rasterization algorithms fall into two main categories, point-sampling and area-sampling. Point-sampling techniques allow for high quality reconstruction, but suffer from aliasing artifacts, time-costly to attenuate. On the other side, area-sampling, popularized by Edwin C. Catmull's Unweighted Area Sampling or UAS, is equivalent to point-sampling at an infinite rate, but reconstruction is restricted to a unit-size box-filter, which offers very poor reconstruction characteristics. We propose a new rasterization algorithm named multiresolution integration, MI which provides high quality reconstruction such as point sampling techniques may do, while achieving speed in the range of unweighted area sampling: a weighted average of box filters is used to approximate the convolution integral between the polygon and any kernel of finite extents. A simple and fast implementation is described, providing high quality 2D rasterization at interactive speed. Examples and benchmarks demonstrate both the quality and speed of this new method.