In
computer graphics, supersampling is the process of
sampling an
image at a
frequency higher than the target
sampling frequency (usually the
pixel frequency). The key idea of supersampling is the same as the one in
oversampling. However, in supersampling, no
analog filtering or
quantization is involved.
Supersampling, when combined with filtered decimation (reducing the sampling frequency by 'averaging down'), is a simple and effective method for anti-aliasing when prefiltering (filtering the continuous signal before any sampling) is difficult or impossible.
Aliasing is a pervasive problem in computer graphics. The sharp edges in computer graphics models represent arbitrarily large frequencies. Furthermore, prefiltering in rendering applications is very difficult. Even for simple lighting models prefiltering requires sophisticated analytical filtering algorithms. For ray-tracing, prefiltering is more or less impossible. Supersampling is a simple and effective alternative in these cases.
Essentially, instead of the normal one sample per pixel, multiple samples per pixel are taken. These multiple samples are arranged on a regular grid which is finer than the pixel grid. The multiple samples for a pixel are then averaged to form a single value for that pixel (this is the filtered decimation 1). For most computer graphics applications, 4 samples per pixel is the minimum, 9 is adequate, and 16 is more than enough.
Supersampling is not perfect. Since there is no analog prefilter, no matter how high the supersampling rate, there'll always be some aliasing present in the resulting image. Monte Carlo integration gets around this limitation by sampling in a random pattern rather than on a regular grid. This has the effect of converting visually objectionable aliasing artifacts to less objectionable random noise.
1 See oversampling for a signal processing view of what happens.