Image Quality Over the Internet

June 16, 2015, SMPTE Entertainment Technology in the Internet Age, Stanford, CA—Pat Griffis from SMPTE talked about the changes coming to TV and other video images. The advent of 4k and 8k sets will change more than the basic picture.

An obvious development in the higher resolution TV sets is the increase in the number of pixels. The more noticeable changes will be in the nature of those pixels that will provide higher spatial resolution, faster moving images, and a wider luminance range. The increased number of dots per inch provides finer and less blocky images, while higher frame rates above the 24 or 30 fps will eliminate a lot of the motion artifacts.

The biggest change will be in the overall brightness, color volume, and dynamic range. Just increasing the pixel count does not provide much difference from HD images. in various tests, most people cannot tell the difference between HD and 4K (3820 x 2160 about 8 megapixels) at the same frame rate and standard dynamic range. Moving to an 8K TV (7680 x 4320 about 33 megapixels) noticeably improves picture quality but may not be worth the large price premium.

Changing the number of bits per pixel increases both dynamic range and color gamut. The nomenclature to define these characteristics start with luminance, which is the luminous intensity of the light source in a given direction waited by the spectral response to the human eye. This parameter is measured in candela per meter squared (cd/m^2) called nits.

Photopic or color vision peaks in the green range near 550 nm. The human eye spectral response drops to 15 percent of this value near 650 nm or red and 475 nm for blue. Luminance levels in nits for black is the minimum scene luminance, diffuse light is a scene reflective white luminance value, and highlights, such as specular reflections and light sources.

Actual levels experienced in the world range from direct sunlight at 1.6 billion to starlight at one millionth nits. The human visual system can adapt across a range from 100 million to one millionth but not simultaneously. The TV standard calls for 100 and it smacks although current televisions deliver from 100 to 500 nits. The cinema standard is 48 nits.

Within a normal viewing environment, people easily experience a luminance dynamic range exceeding 10,000 nits. For example a dark carpet in an unlit room might be 0.5 while the light in the hallway is 6000 nits. A room with a window could have dark areas at 0.08 nits while the sunlit outdoor scene with the over 6000. In an outdoor environment specular highlights could exceed 330,000 nits while shadows would be at 40. These ranges do not have to be in a large area a flower could have a dark central area at 188 nits while a bright colored petal next to it could be at 14,700.

In addition to calling for higher dynamic range, truer visual experience require larger color volumes. See figure 1. Rec.709 defines a color space much smaller than human vision, which places many colors outside of the displayable color volume. Emissive colors can be bright and very saturated.

The entertainment dynamic range must change to encompass a greater portion of the human visual system. Tests showed that 84 percent of viewers are satisfied and comfortable with a dynamic range from 0.005 to 10,000 nits. The bottom under the range defines black, which is an elusive target. Reference video calls for 0.01 nit for video and cinema even though the best consumer devices can achieve half of that number. The visual system limit is in the range of 0.00001-0.000001 nit even though the: threshold is about 0.003. With a long of adaptation time you can see a handful of photons.

At the other extreme, white is defined as 80 to 100 nits for reference video and 48 for cinema. The brightest consumer devices are running about 1500 nits and commercial devices can achieve 5000. However, 10,000 nits is easy look at and measure and still are not as bright as specular highlights in the real world.

The current electro-optic transfer function standard BT-1886 is based on gamma and is used up to 100 nits. Practical future systems will need between 10 and 12 bits for sufficient precision, but a change to a human visual system model would make this level of precision close to actual human vision, based on the best reference work by Barten. See figure 2.

The Barten ramp shows increasing compression with higher levels of luminance to define step thresholds. Values below the threshold are invisible on provide smooth gradients, while values above the threshold exhibit visible artifacts. A new standard ST-2084 uses a perceptual quantizer (PQ) model that is a hybrid of gamma at low luminance levels and log at higher levels. A 12-bit PQ maps closely to the Barten ramp. From 0 to 1 nit is similar to a 15-bit gamma and above 10 is close to a 13-bit log. This curve is most efficient use of bits throughout the entire range of precision and all values are below the threshold of visibility.