Game Development Reference
In-Depth Information
Second-Depth Antialiasing
Emil Persson
Introduction and Previous Work
For nearly a decade multisampling was the one and only antialiasing solution for
real-time rendering. For most of this time, multisampling was never challenged
because it worked relatively well. As rendering technology has developed and
engines become more advanced, multisampling has become an increasingly big-
ger stumbling block and maintenance problem when developing new rendering
techniques. With the recent popularization of deferred shading, the memory in-
eciency of multisample antialiasing (MSAA) has further been magnified to a
point where it becomes impractical, in particular on current generation consoles,
and the inherent weaknesses of this technique become much more apparent.
In recent years a number of interesting alternatives to multisampling has
emerged. In 2009 morphological antialiasing (MLAA) [Reshetov 09] was intro-
duced, starting off a wave of new techniques. The fundamental difference between
MLAA and MSAA is that the former is entirely a post-process operation. This
means that the algorithm is provided a finished rendered image, which it then an-
alyzes and antialiases. This works by first detecting edges in the image (defined,
for instance, by a certain difference in luminance between neighboring pixels)
and then classifying those edges into a set of shapes. From these shapes it re-
constructs what the original edge might have been and uses that to compute a
coverage value used for blending with neighboring pixels. MLAA was initially
a CPU-based technique for ray tracers but was later expanded to GPU imple-
mentations by others [Jimenez et al. 11]. At the point of this writing, several
games have shipped with MLAA in some form as their antialiasing solution. An-
other very popular technique is fast approximate antialiasing (FXAA) [Lottes 11]
that, due to its simple single-pass implementation with optimized targets for all
important platforms, is a very convenient alternative.
The above and several other approaches can be described as post-process an-
tialiasing techniques, because they require no particular knowledge about how the
image was generated. Being decoupled from scene rendering is a great advantage,
Search Nedrilad ::

Custom Search