Lost in Translation: Going from TV to PC
Y, R-Y, and B-Y are converted to RGB and vice versa via matrix math translation. There are a number of sets of matrix coefficients used to accomplish this. Figure 2 lists the most frequently used sets. (Note that R, G, and B are gamma-corrected values; gamma will be discussed shortly.)
Figure 2. Color conversion matrix coefficients for digital and analog television
Since matrix-coefficient values applied at the broadcast encoder vary depending on the source of the original content, this leads to the logical query: Are the correct broadcast matrix coefficients being used during processing in the PC?
Figure 2 shows the difference between matrix coefficients for digital and analog television. To add to the challenge, these are not the only sets of color space conversion coefficients. The inverse values are used to convert from Y, R-Y, and B-Y to RGB. Get the color conversion matrix wrong and color fidelity suffers.
Another issue that can lead to loss of resolution is the relationship between luminance and chrominance sampling areas. Color sampling for RGB is 1:1:1, although it’s referred to as 4:4:4 in TV engineering jargon, and there is a one-to-one correlation between RGB values. Because the eye is less sensitive to chrominance than it is to luminance, during the conversion to Y, R-Y, and B-Y, professional production systems used in compressed content workflows employ a 4:2:2 color sampling space, while MPEG relies on 4:2:0 to further reduce data for transmission. Figure 3 compares the techniques.
Figure 3. Comparison of color-sampling techniques and display area
When delivering TV content over the web, compression encoders can use any of a variety of color sampling spaces. Try to decode a 4:1:1 color space using 4:2:0 methods and quality takes a hit, again.
Another fundamental difference between DTV systems and computer graphics systems is the number and range of quantization levels for RGB or Y, R-Y, and B-Y information.
Computer systems represent RGB with a full range (Figure 4). For 8-bit color, the range is 0–255. ATSC DTV, due to the origins of Advanced TV as a hybrid analog/digital technology, restricts this range to 16–235. The values less than 16 are present in order to enable digital encoding of sync signals. This innate difference raises havoc with the TV production chain, and color "legalization" is a required step when computer-generated graphics are combined with video before it hits the air.
Figure 4. Color representation ranges
ATSC full range results in 220 levels as compared to the 256 used by PC systems. That’s a 15% reduction in color resolution. Perhaps worse is the impact of round-off errors that result when scaling TV color quantization levels to PC proportions. Cumulative errors in computational accuracy are a big problem.
Color Primaries and Gamut
All display technologies rely on some physical means to produce light. Broadcast standards-setting bodies, such as the Society of Motion Picture and Television Engineers (SMPTE) and the International Telecommunications Union (ITU), have defined the spectral characteristics of the primary colors (red, green, and blue; RGB) that are used in imaging devices. Both ITU 601 and ITU 709 are based on CRT technology. When was the last time you saw a CRT used for a PC display?
As Figure 5 (left) shows, when the coordinates of the RGB color primaries are plotted on the CIE color chart, triangular color spaces are defined. Not only will different colors be produced by differing sets of RGB values, but there are regions that can be produced with one color primary set but not with another.
Figure 5. Color gamut: Given the same input value, different colors are produced.
Reach out to an international audience. Here are video translation steps to follow for strong results, as well as expert tips for high quality.