>Psychtoolbox>PsychDocumentation

AntiAliasing – Psychtoolbox facilities for Anti-Aliasing.

According to http://www.wikipedia.org , anti-aliasing is the
technique of minimizing aliasing (jagged or blocky patterns)
when representing a high-resolution signal at a lower resolution.

If you want to know in detail about the reason for aliasing artifacts
in digital computer images and the theoretical approaches for removing
or reducing it, as a starter, search and read on Wikipedia about “Aliasing”
and “Anti-Aliasing”, or read a book or website about computer graphics
or digital image processing.

Psychtoolbox allows for multiple approaches to anti-alias your stimuli.
The methods all represent some tradeoff between ease of use, generality
and amount of brute-force applied to implement them.

Multisampling / Supersampling:

The easiest (for you) and at the same time most general approach is the
brute-force approach, called “Multisampling” or “Supersampling”, also
known as “Full scene anti-aliasing” or FSAA. You can enable multisampling
for an onscreen window by simply setting the optional argument ‘multisample’
of the Screen(‘OpenWindow’, …) - subfunction to a value greater than zero.
Psychtoolbox will check if your graphics hardware supports multisampling and
enable it for the new onscreen window, if possible. In a nutshell, the graphics
hardware will then perform all drawing operations (including all images, all 2D shapes
like points, discs, lines, rectangles and polygons, text and all 3D OpenGL rendering)
at a much higher resolution than your display resolution, computing multiple color
samples for each output pixel. When calling Screen(‘Flip’), the hardware will
downsample this high resolution image by use of a proper lowpass filter, combining
the multiple color values of the multiple samples per pixel into a single color
value for the final output pixel. A typical algorithm for this would be a weighted
average of the color values of the multiple samples, but depending on your gfx-
hardware, the algorithm may use more clever schemes. The value of the ‘multisample’
parameter roughly specifies, how many samples should be used for each final pixel.
A higher value means a higher quality, but at the same time higher requirements in
terms of video memory usage and processing time.

Psychtoolbox tries to select a value as close as possible to the requested value: It
rounds up to the closest value supported by your hardware (e.g., if you request 6
samples, but the hardware only supports either 4 or 8 samples, PTB will select 8
samples). If you request a value that is greater than the highest supported value,
PTB will use the maximum value supported by your system.

The algorithms used for multisampling differ between each model of graphics hardware:
NVidia’s gfx-cards do it differently than ATI’s cards and each model from NVidia or
ATI also does it differently from its predecessors. For a general overview of how
multisampling and supersampling works, have a look at Wikipedias articles on
“Supersampling”. For specific gfx-cards, the hardware vendors provide white papers
and specifications on how their devices operate at specific settings, e.g., one
presentation of NVidia is http://developer.nvidia.com/object/gdc_ogl_multisample.html
A more detailed article is http://www.3dcenter.org/artikel/geforcefx_aa_modi/index_english.php

As stated above, this brute-force approach of anti-aliasing requires minimal
work for you (setting one optional parameter) and it works always, but it is
expensive. For a multisample value of n, your hardware will need roughly n+1 times
the amount of video memory and it will need more than n times the amount of computation
time for drawing and showing your stimulus. The maximum value that you can set for n
therefore not only depends on how modern your graphics-card is, but is also limited
by the amount of VRAM installed on your gfx-card, the screen resolution you’ve selected,
the complexity of your stimulus and the stimulus update rate you need, so choose a
reasonable tradeoff for your purpose.

Some sort of multisampling/supersampling should be supported by basically any piece
of graphics hardware for Intel PC’s under M$-Windows or GNU/Linux sold by NVidia or ATI
after beginning of 2001. On MacOS-X, all supported gfx-adapters, except the ATI Rage 128
and ATI Rage 128 Pro, should allow for at least 4-way sampling.

Typical values on todays hardware: GeForce-3 series: 4-way sampling,
GeForce 7800 series: 16 way multisampling.

To be continued…