Monday, December 19, 2005

"The perils of colour-space conversion" - an article for the Root6 Dr. Watson newsletter

At work we produce a newsletter which I sometimes do a technical piece for - so forgive the slightly client facing tone of this piece!

An important aspect of the production/post-production chain is maintaining correct colourimetry. If the director of photography or the lighting-cameraman want that certain shade of red to be correctly delivered to the viewer then attention needs to be paid to the correct representation from the camera (be it standard or high definition or even film) through all transfer operations (potentially going between resolutions, YUV/RGB colour spaces and bit depths) to the final display surface (be it a CRT, LCD or even cinema screen). In truth colour-space management for film is a complex issue best handled by specialists like Filmlight (who are represented in the UK by Root6) and is perhaps beyond the scope of some technical notes in a newsletter! That said there are many points worth making if you are acquiring or delivering for high definition television and worried about going between colour spaces.

The CIE Chromaticity diagram (first published in 1931!) shows the gamut of human vision – essentially any display surface is a subset of this diagram and will be a triangle with red, green and blue apexes and white (actually monochrome – as the luminance of the image is reduced it tends through grey to black) in the centre. In the case of “illuminant D” (AKA “D6500” or “EBU phosphors”) – the standard definition colour standard used since the sixties in Europe we enjoy a slightly wider red range than our colonial cousins but every gamut (television, film or print) is a poor compromise on what your eyes can handle. This is where the problem begins – you have a very critical instrument at your disposal to see these differences.

Part of the problem is that all of our machines acquire images in the RGB space (TV cameras, Telecines, graphics workstations etc.) but for the most part we post-produce in a YUV space (with the exception of Sony’s new HDCam SR format, an RGB high-definition VTR) which represents an immediate lowering of the colour space. This has been the case for a long time and is well understood. Manufacturers have agreed a common “matrix” for transcoding. To make the luminance portion of the component signal the following is used:

y = 0.299 * r + 0.587 * g + 0.114 * b

Well, this is the case for standard definition (AKA “601”), but for high-def (AKA “709”):

y = 0.213 * r + 0.715 * g + 0.072 * b

Which, even if you’re not so into the maths, will give different values for the luminance (the overall level and hence look of the picture). This makes it doubly important that you get your cameraman to record some colour bars at the head of each rushes tape and that your editor checks alignment on his scope before he starts adding captions etc. The best of breed digital picture instruments are from Tektronics who Root6 are pleased to represent.

If you find you have a colour space issue (particularly going between standard and high definition formats) then the Belle Nuit Montage test chart is a good starting point. It can highlight all the common transcoding errors – be it the limited 8-bit range of old D1 videotape to some of the sub-sampling issues associated with HDCam. You can download the file at various resolutions from their website and by injecting it at the start of your workflow any inadequacies are quickly revealed and can be corrected.

No comments: