Anti-aliasing Filters & Sampling


Background Information

All CGI images are made of fundamental building blocks called pixels. a pixel is the smallest breakdown of an image, and can only be made of one one color. When an image is rendered, the renderer needs to assign a single color to each pixel of the overall image depending on what object that pixel represents. This standard process brings up some problem as to what we call aliasing. where thin lines or diagonal lines are present in an image. To overcome this entire thing, we use a process called Anti-Aliasing. Here each pixel is multi-sampled i.e each pixel is broken down into multiples smaller parts which equals to more color information and then all the samples are averaged out using a Anti-alias filter, so as to obtain the final one color per pixel value.

Sampling Methods

Fixed Sampling - Uses a fixed number of samples per pixel when processing an image.

Adaptive Sampling - The number of samples used per pixel varies depending on the contrast of your scene. Either vary by a fixed distance

Custom Sampling - Custom Sampling allows you to tune the Min Sample Level and Max Sample Level independently, while retaining true adaptive sampling.

Anti-aliasing Filters


Box Filter - A pretty Balanced Speed/Quality Filter. Gets you the fastest results, Due to it having a Step Curve the sample contributions are always taken from the center of the pixel.

Area Filter - A little more advanced than the Box Filter in the case that it blur the image slighty.

Triangle/Tent Filter - Overall SlowerSpeed/Better Quality than the box filter. In this case samples at the center have maximum contribution while samples further away from the center have a leanear falloff to the contribution.

Gaussian Filter - Produces Best Quality/Slowest Speed. Gauss uses a curved fall-off for determining sample contributions. The contribution falloff from the center to the edge of the pixel is smooth. Gauss requires a higher filter size than normal due to which the images get a little blurry.

Mitchell & Lanczos - Both are variants to the Gauss, but they tend to increase the overall contrast and the sharpness of the image. A good allround filter that gives reasonable sharpness

Quadratic B-Spline Filter - Produces Good Quality at speeds faster than the Gauss Filter. Compared to the Gauss filter the contribution from the center of the pixel does not peak but archieves a more quadratic curve.

CatmullRom Filter - Speed Same as Mitchell and Lanczos  but Catmull-Rom filter, gives the most sharpening. the contribution curve travels also in negative to give best overall contribution.

Sinc Filter - i dont know how this works.. if someone find out please let me know