Coverage Mask?

Blurring

Another issue with anti-aliasing is that of blurring. While there has been some ongoing controversy about this issue, it is understandable why some complain about blurring. What one must consider first though, is what blurring actually is. Many processes in computer graphics can be considered blurring, including basic technologies such as bilinear filtering. To say something blurs does not necessarily make it a bad thing. By filtering, you simply remove those high-range signals that normally would not be able to be displayed accurately to begin with.

When working in lower resolutions, you can lack sufficient resolution to display needed texture detail. With super-sampling, you are averaging four input pixels to achieve your single output pixel. This usually leads to each pixel having a more accurate value, but sometimes, at low screen resolutions, there is not sufficient base accuracy (resolution) to display all the necessary detail, resulting in an image that some describe as being blurred.

This can be helped to some extent by increasing the mip-map levels to a higher detail (LOD bias). By doing so, you start out with greater initial detail and your result is therefore less "blurred". Basically, this scenario is one in which your video adapter is trying to display too much accuracy without sufficient resolution.

Other types?

There are a variety of other implementations for anti-aliasing aside from super-sampling, but typically they revolve around edge anti-aliasing and texture filtering. One such implementation of edge anti-aliasing involves use of a coverage mask. While there are different methods to achieve edge anti-aliasing, a coverage mask implementation can offer some of the best available quality.

A coverage mask is a grid placed over a pixel or group of pixels to separate them into sub-pixels. This is done by breaking scanlines down into subscanlines, which are then segmented. From this a triangle edge intersection with scanlines is determined and sub-pixels are calculated based on the intersection with the subscanlines and segments. In doing this you locate the edge of a triangle and use your coverage mask to determine which sub-samples will fall onto the triangle edge.

By doing this you determine your coverage and you are able to calculate the intensity of each pixel along the edge. This results in edge anti-aliasing. However, an implementation such as this is often considered rather tricky and costly to implement.

Staggering

It is interesting to note that image quality can be improved with a coverage mask in a way similar to just how a rotated grid improves quality over an ordered grid. To do this, a technique known as staggering can used. This is done by generating a high quality (32 or 64-sample) coverage mask, but reducing the performance penalty by using only half of the sub-samples and alternating positions within the mask.

Here are two illustrations. The first shows a standard 16-sample coverage mask and the second shows the mask, but with staggering. Notice that while the staggered image only has half the samples, each sample is offset so that is able to achieve quality considerably closer to the normal mask.

There really is a great deal more involved in edge anti-aliasing than we have time to address in this article. There are a variety of implementations, such as Z3 and others, and each could get an article of its own. We will focus, however, on one implementation in particular, multi-sampling.