Wednesday, June 4, 2014

Distribution vs. Contribution Encoding: Part 2

Continuing the discussion of source vs. contribution encoders we began last time, we indicated that, while encoding essentially involves discarding some of the pixels in an image, source encoders typically retain twice the number of pixels for each frame of video as a distribution encoder. In industry jargon, source encoders are said to operate in 4:2:2 mode while distribution encoders operate in 4:2:0 mode.

The other difference we mentioned between source and distribution encoders, besides the number of pixels retained in an encoded image, is the number of bits used to denote colors in a pixel, known as “8-bit” color, “10-bit” color, and so on.

As the name implies, 8-bit color uses 8 bits to denote each of the three primary colors used to compose pixels: 8 bits for red, 8 bits for green, and 8 bits for blue – 24-bits in total for each pixel, or enough bits to distinguish 256 distinct shades of each primary color. Since any shade of one color can be combined with any shade of each of the other two colors, the total number of possible colors with 8 bit encoding is 256 x 256 x 256 = 16,777,216 possible colors. Since, even under optimal viewing conditions, the human eye can distinguish fewer than 10 million colors, 8-bit color is more than adequate to reproduce every color we can possibly see. For this reason, 8-bit color (24-bit pixels) is also known as “true color”.

But that is on the viewing or distribution side. On the source or contribution side, just as it is often desirable to retain twice the number of pixels needed to accurately reconstruct an image, it is often desirable to have significantly higher color resolution than needed for viewing. By allowing 10 bits to denote each primary color, the number of primary shades quadruples, from 256 to 1024, providing a total pool of 1024 x 1024 x 1024 = 1,073,741,824 colors.

While having a billion colors available does nothing to enhance the viewing experience – the human eye is simply incapable of resolving the tiny differences between the hundred nearest shades in a billion color pallet – it can and does help on the source side, when video is edited. That is because many editing steps are “lossy”, which is to say, the edited images contain less information than the original images. By starting with a billion color pallet, however, the editing losses are generally imperceptible. Reducing a billion color pallet to millions of color does not degrade the viewing experience, since the hundreds of fine distinctions lost are imperceptible to the eye. Whereas reducing a 16 million color pallet to thousands of colors does result in a clearly perceptible loss of color fidelity.

The same argument holds for the desirability of starting the editing process with twice as many pixels as actually needed for viewing. Losing half the pixels from a 4:2:2 image does not result in any significant loss of image quality. Whereas losing half the pixels from a 4:2:0 image does noticeably degrade the image.

At Telairity we support both distribution and contribution encoders and all our designed and manufactured in the US. Visit our website to learn more.