<img height="1" width="1" src="https://www.facebook.com/tr?id=414634002484912&amp;ev=PageView &amp;noscript=1">

High Dynamic Range Intermediate

October 23, 2014

Gary Demos, CEO/founder of Image Essence LLC, which is developing wide-dynamic-range codec technology based upon a combination of wavelets, optimal filters, and flowfields, spoke about the HDR intermediate. Historically, the properties of the Cathode Ray Tube dominated video signal attributes, providing a 300:1 dynamic range. The current system is Rec709 but is still tied to phosphors. In 2003, when flat panels began to replace CRTs, they offered brightness, color gamut and dynamic range that greatly surpassed CRTs. "We've had a long time of TV color gamuts defined as xy Chromaticites," said Demos. "But we have a wildly different spectrums."

The accuracy of colors can be summarized on the following guidelines: narrow gamuts will be more accurate (within that gamut); more than three primaries will be more accurate; broadcer primaries will be more accurate. Color matching functions of individuals are more accurate than the generic one-size-fits-all CIE 1931. Generic classifications of individuals by age can improve color matching accuracy somewhat.

"We also have a gamut reduction issue," said Demos. "When we author HDR, I'm thinking of what we did will LAD. If we think of the LAD as near gray, the blacks will float a little bit."  How should the signal to displays/projectors be optimized? "Trust the carefully determined 12-bit 48nit gamma 2.6 used in DCinema and trust the pure logarithm brighter than 48nits where self-relative perception becomes 0.2 percent and remains that going brighter," he said, noting that the solution is to use 16-bit half-float RGB."

Color appearance affects include ambient light within the room, perceived colorfulness and many other factors. "I suggest we use linear light (gamma 1.0)," he said. He showed a chart showing a combination of Rec709, P3, BT2020 and cyan. "In the current ecosystem of displays we send Rec709 to everybody," he said. "The right answer is to put out a signal that is bigger than the biggest display you want to support and let each display use as much of that space as it's capable of using."

"I like to build on simple principles," he said. "I like red = green = blue = white. I like finite slope for all scene brightness levels when transforming to all HDR intermediate master brightness levels. You can have an absolute black. Wider gamuts are preferable over negative RGB values, and I recommend use of 16-bit half-float (32-bit floats for computation). If mastering on a low dynamic range display or projector, use reduced contrast and/or desaturated gamut."

He also pointed out many issues with negative numbers. "I don't know any convincing story of what we do with this," he said. "It's a tricky problem. I think we should have negative values at least for calibration." Another way of working which is to apply an aesthetic rendering. If you start with an aesthetic rendering, you get part of the way there," he said. "That doesn't mean we don't need to dial it in with a DI system. But I also think it's useful to have some other simplified variants such as tone curve. Dealing with face-tone specific HDR rendering is also critical. Things have to be done to warm the shadows, etc., and these are tricky things."

He recommended two kinds of correction: the scene-referred one, which is potentially alias-free, contains all available gamut and range info and has a simplified relationship to aesthetic rendering. An HDR intermediate is output referred and device independent. It may have been color-graded and may have reduced gamut and range. "It may also be associated with a specific reference display or projector and may contain some aliasing on sharp edges."

Among his other conclusions, Demos said that  one or more HDR Intermediates can embody creative intent and LAD mid-gray and mid-white can anchor HDR scenes.

Tag(s): HDR

Debra Kaufman

Related Posts