Re [PD] pix_freeframe

[pär F]

Sorry about the double-post, wrong key sequence.




I kind of get the feeling this thread is in need of some  
clarification regarding RGB, YUV, 4-1-1, 4-2-0, 4-2-2 and 4-4-4. I'll  
threat you all as idiots and try to explain it from scratch.

A digital image is built up by pixels.

If the picture is in gray-scale you only need one number to define  
the tone of each pixel. The defining number is somewhere in a  
predefined range, the greater the range the more possible tones. If  
the range is only 0 and 1 you can define only black and white. If the  
range consists of more than about 200 possible variations you can  
define a seemingly unbroken grayscale between black and white (8 bits  
equals 256 variations). If you need to make vast manipulations of the  
tones of your pixels you might need an even greater range of possible  
tones in your original pixels (that's why there is such things as 16  
bit and 32 bit grayscale images).

In a RGB-format image the color is defined by 3 grayscale channels  
(that is three separate grayscale images really), one for red, one  
for green and one for blue. It is also possible to ad other channels  
to an image, such as alpha-channels for transparency or z-channels  
for depth or any other channel you find useful. Every pixel in your  
image could be defined by 1, 3, 4, 5 or even more channels, each  
consisting of 1, 2, 4, 8, 10, 16, or 32 bits.

YUV also contains 3 channels, but the luminance (brightness)  
information (Y) of the image is separated from the color information  
(u and v). The human brain is much less sensitive for changes in  
color than for changes in luminance. Therefore if the color  
information is separated from the luminance you can save space by  
removing some of the information in the color-channels while  
retaining approximately the same visual qualities. 4-1-1 means you  
have only one pixel in every color channel per pixel in the luminance- 
channel (4-2-0 is a variant where the color-pixels alternate between  
u and v every frame (or field?), the compression is the same). 4-2-2  
means 1 color-channel pixel per 2 luminance pixels. In this respect  
RGB is actually always 4-4-4.


4-2-2 or 4-1-1 reduces space in a great way, but if you need to make  
manipulations that moves information from color to luminance the  
method might be a rather bad idéa. Worse even, if you need to use  
color-information for creating an alpha-channel (blue-screen/green- 
screen f.ex.) you might experience serious loss of quality.


regards, Pär