Alpha Channel

From RMIT Visual Effects
Jump to: navigation, search

A digital image is composed of red, green and blue (RGB) information. These are stored in three separate channels within the same image file. The neat thing is that any sort of information can also be stored in such a channel, and is used frequently in multi pass rendering. The most common form of extra information is transparency. This is stored in the alpha channel. In the act of creating this transparency, the alpha channel becomes the matte (which is a very old term that comes from the pre-digital age).

The first person to use a channel in such a way was Alvy Ray Smith. He was cofounder of Pixar. This approach was further described in a paper by Thomas_Porter (Pixar) and Tom Duff (1). In this paper they described a process in which an alpha channel might be used to create the illusion of transparency in a compositing operation. In order for the alpha to act as a matte, it needs to be premultiplied.

Alphas, Mattes, Masks and Premultiplication? Whats the difference? Summary below:

Alphas, Mattes, Masks and Premultiplication
Name Description
Alpha A 'spare' channel, embedded into an RGB image. It is commonly used as a matte.
Matte In the case of a foreground being laid (composited) over a background, this is the channel that creates the impression of selective transparency. Sometimes the alpha channel of the foreground is used for this purpose. Sometimes an alpha is acquired from other sources.
Mask This is a channel used to moderate an adjustment or a filter. In Nuke, this is usually fed in from the right hand side input.
Premultiplication In order for the alpha to act as a matte, it needs to be premultiplied.
Layers Charmingly, Nuke has yet another name up its sleeve for channel information: layers. A layer is a collection of channels, typically from a 3D multi pass render. I won't cover this in detail in my wiki, but more info is here for those who are interested.

References

(1) Porter, Thomas; Tom Duff (1984). "Compositing Digital Images". Computer Graphics 18 (3): 253–259. doi:10.1145/800031.808606. ISBN 0-89791-138-5.