Jonas Notes

Aliasing

2 min read

Key Concepts in Aliasing and Antialiasing

  1. Aliasing in Computer Graphics:

    • Types of Aliasing: Occurs in three main areas:
      • Pixels: Jagged edges or “staircase” effects in rendered images.
      • Textures: Moiré patterns or other distortions when viewing textured surfaces.
      • Time: Artifacts like motion strobe effects when sampling at discrete time intervals.

  2. Filtering Techniques:

    • Filters help in reconstructing images from samples to reduce aliasing. Common types include:
      • Box Filter: Simple and fast but may produce blocky artifacts.
      • Tent Filter: Averages over a larger area, giving smoother results than the box filter.
      • Sinc Filter: Theoretically ideal for reconstruction but impractical due to its infinite extent and negative lobes.

  3. Sampling Schemes:

    • Various sampling patterns help in reducing aliasing:
      • Quincunx: Uses a pattern of 5 sample points, effectively averaging 2 samples per pixel while providing smooth antialiasing.
      • Grid and Rotated Grid Super Sampling (RGSS): Offsets sample points to minimize regular pattern artifacts.
      • Checker and 8-Rooks Patterns: Unique patterns that distribute samples in non-regular ways to reduce aliasing.

  4. Jittering:

    • How it Works: Slightly randomizes sample positions to reduce regular aliasing artifacts, replacing them with noise, which is less noticeable.
    • Effect on Perception: Humans generally prefer the noise produced by jittering over regular aliasing artifacts, as it appears more natural.

  5. Antialiasing Techniques:

    • Supersampling: Takes multiple samples per pixel, running the fragment shader for each sample. This technique is computationally expensive but highly effective.
    • Multisampling (MSAA): Reduces computation by sharing shading information across samples, saving time by only executing the fragment shader once per pixel.
    • Coverage Sampling (CSAA): Further optimizes by using index buffers, storing color and depth once per fragment rather than for each sample.

  6. Quincunx Pattern and Weights:

    • Quincunx uses five sample points per pixel (center and four corners) with specific weights, averaging to two samples per pixel in cost. It balances quality and performance efficiently.