About Adjusting Sampling to Improve Image Quality

You can use several standard render presets to remove the jagged effect from the display of diagonal lines and curved edges.

Because the image on a monitor is made up of discrete pixels on a fixed grid, scenes rendered using the Draft or Low render presets produce jagged and inaccurate images.

In the following example, image quality is highly degraded - Min. Samples = 1/64; Max. Samples = 1/4.

With a higher resolution (and thus smaller pixels), aliasing is less apparent. However, you can reduce the effect further by using anti-aliasing techniques. The anti-aliasing technique used by the renderer is called sampling. Sampling is controlled by the sample and filter settings on the Advanced Render Settings palette.

Sampling provides a "best guess" color for each rendered pixel. The renderer first samples the scene color at locations within the pixel or along the pixel's edge, then uses a filter to combine the samples into a single pixel color. Increasing the minimum and maximum sample ranges greatly improves the quality of a rendering.

In the following example, increased sampling smooths edges - Min. Samples = 1; Max. Samples = 16.

There are five filter methods: Box, Gauss, Triangle, Mitchell, or Lanczos. Box, the default, is the quickest. Mitchell is often the most accurate. The Box filter combines samples evenly, without weighting them. Each of the other filters uses a particular curve to weight samples before combining them.

Anti-aliasing techniques involve at least two trade-offs.

You have to decide how much time you want to spend rendering and how good you want the final rendering to be. These trade-offs depend partly on the hardware you're using and partly on the audience for your work.

The standard render presets offer five levels of anti-aliasing. Draft is set to use the lowest level of anti-aliasing, but it provides the fastest rendering speed. Each successive level is slower to compute. The anti-aliasing algorithm makes the process more efficient by allowing the program to decide for any given pixel that fewer samples need to be computed.