Luminance, luma, and the migration to DTV

Presented at the 32nd SMPTE
Advanced Motion Imaging Conference,

Toronto, Feb. 6, 1998


Since 1953, we have been using the wrong block diagram for color video! The principles of color science dictate that we mix linear RGB to make true luminance, denoted Y. This is known as the Principle of Constant Luminance. But in video we depart from that principle, and implement an engineering approximation: We mix nonlinear ("gamma corrected") R'G'B' to make what I call luma, denoted Y'. (Many video engineers carelessly call this luminance.) To form luma, we use the theoretical coefficients of color science, but we use them in the wrong block diagram: We apply gamma correction before the mixing, instead of after. This alteration in the block diagram is more or less inconsequential in practice, though the departure from theory is apparent in the dark band seen between the green and magenta color bars of the standard video test pattern.

The Rec. 709 HDTV standard has, wrongly in my opinion, adopted a new set of luma coefficients. There is an incremental benefit in using the "right" coefficients, even in the wrong block diagram. The change would be sensible if HDTV was a closed system. But changing the coefficients introduces a second flavor of Y'CBCR - effectively, there is now one flavor of Y'CBCR for small pictures (SDTV), and a different flavor of Y'CBCR for big pictures (HDTV).

HDTV originated in the studio will be downconverted to SDTV for simulcast. Consumer ATV receivers will display conventional video sources through upconversion. Even studios are likely to use upconversion, to allow the conventional studio equipment as a cheap way to originate "HDTV" in the early phases of deployment. No studio upconverters or downconverters today do the necessary processing to accommodate the different flavors of luma coefficients - instead, production personnel are faced with having to perform manual colour correction. In consumer equipment, it seems highly unlikely that the correction will ever be properly implemented in hardware or software: It is impractical to perform a 3 x 3 matrix multiplication for every pixel, especially when the DCT of the decompression requires only 3 multiplies per pixel! Instead, the consumer will be faced with poor colour reproduction.

The computer and communications industries find it hard to understand why color coding parameters should change as a function of image size. The different flavor means that it is impossible to cut and paste Y'CBCR data from a big image to a little one.

The MPEG-2 standard provides, in its data stream, an indication of the luma coefficient set (flavor) used in encoding. I advocate adoption of a recommended practice for ATV that calls for the use of Rec. 601 luma coefficients - in other words, that HDTV, ATV, and DTV use the same flavor of luma coefficients that have been used since 1953. This will aid the transition to high definition, ATV, and DTV, by delivering the high colour quality that studio engineers, program producers, and consumers expect.

No printed version of this paper is available at the moment, though I plan to write it up eventually and submit it for publication. An audiotape recording of the conference presentation is available from SMPTE.

See also, Constant Luminance.

Charles Poynton
- Video engineering