ETIA 2015: Introduction to Image Quality

Our first morning kicked off with a pair of talks dedicated to setting the stage of what is needed to render high-quality images for entertainment purposes. Dolby's Pat Griffis started off with a detailed look at both the number of pixels needed per frame, but more importantly how we can measure and improve the quality of those pixels through increasing the dynamic range in both luminance and color -- e.g. the color volume -- of the display. He showed some compelling examples of how large-gamut displays like those developed by Dolby can provide a more-realistic rendering of natural scenes, particularly with bright, saturated colors.

Griffis proposed that a useful target dynamic range for the future of TV would be .005 to 10K nits. That goes darker than the current cinema standard for black, but not as low as an adapted human eye, so SMPTE is working with a nominal minimum of 0 nits. On the bright end of the range, while natural specular highlights can be much brighter than 10,000 nits, that's an easy number to measure, and is beyond what consumer and professional devices can do today -- making it the favored choice for a target today. Griffis explained how this massive new brightness range renders traditional gamma as very inefficient, which was a great segue to the next panel on replacing gamma with a modern engineered alternative. During the question period, Conference Chair Joyce Farrell pointed out that compression issues would be even more severe when we start using lightfield displays. In response to a question about how much bit-depth is enough, Griffis said that for most real world video 10-bits is probably enough, while for computer-generated content -- which is noise-free -- you need 12-bits.

Adobe's Lars Borg and Dolby's Scott Miller followed with a panel on PQ -- a replacement for gamma based on the way the human visual system perceives changes in luminance. In particular, this is critical as entertainment devices become much brighter than the 100 nits that was possible with CRT displays. Miller started by taking us through the need for high-dynamic-range displays, and showed that with a high-quality test setup like the 20,000 nit Dolby test environment, users were enthusiastic about the additional brightness and contrast available. From the need to represent a large range, like the 0 to 10,000 nits being used in SMPTE's work, Miller used Barten's work in human contrast observation to build a new encoding curve that provides a more effective mapping than the current Gamma encoding, or even previous attempts to extend Gamma to 10 or 12-bits. PQ (SMPTE ST 2084) is a 12-bit curve fit to the Barten observations, designed to operate well -- meaning that it is visually transparent -- from 0 to 10,000 nits, that also works well at 10-bits for existing systems.

Borg followed on with a discussion of where PQ can be used in color workflows, including capture of compressed video, and potentially in grading and distribution over time. Challenges on the distribution side include a lack of standards for HDR TV signals and the variance between HDR displays. Producing a good signal for the reduced dynamic range of conventional SDR TVs is also an important element of adopting PQ. Unfortunately, this transform is content-dependent, making it a complex process, and an active research area.