High Dynamic Range for HD and Adaptive Bitrate Streaming (2017)

By Sean T. McCarthy, Ph.D., Sean McCarthy, Ph.D. Consulting

When High Dynamic Range (HDR) began to emerge a few years ago as a great new television experience, it was almost invariably thought of as only one part of Ultra HD/4k along with a wider range of colors and higher bit depths for video coding. Now though, many are wondering if HDR can be an independent contributor that makes for better TV experiences even without higher 4k resolutions. Indeed, several MVPDs, broadcasters and industry associations are moving towards enhanced programming and distribution that marries HDR with HD and sub-HD resolutions. The next-generation television standard, ATSC 3.0, for example, enables distribution of HDR and Wide Color Gamut (WCG) content at any resolution. Similarly, leading internet-based streaming services are leveraging multi-resolution adaptive bitrate (ABR) protocols to deliver HDR experiences.

The potential wrinkle is that HDR is still often tied to Ultra HD/4k/10-bit in the content creation studios and production houses. The HDR highlights that make HDR shine are often very small and localized.

Thus, converting original Ultra HD HDR content to lower resolutions for HD and ABR streaming runs the risk of obliterating those visually potent HDR highlights. When the resulting video is compressed with High-Efficiency Video Coding (HEVC), HDR distortions could be made worse, particularly for aggressive streaming profiles.

In this paper, we use a recently developed method for measuring distortion in HDR video to quantify the impact of encoding original Ultra HD HDR content at HD and ABR resolutions. Our method is intrinsically independent of any particular HDR transfer function and may thus be applied to HDR content having Hybrid-Log Gamma (HLG), Perceptual Quantizer (PQ), or other transfer characteristic.

Specifically, we show:

  1. That each frame of HDR video can be represented as the sum of a Spatial Detail signal and a Foundation Image. The Spatial Detail signal contains the localized contrast variations associated with HDR features. The Foundation Image contains the more smoothly varying large-area contrast variations associated with textures and backgrounds.
  2. The Spatial Detail signal is the main contributor to HDR distortions introduced by rescaling and compression. The Foundation Image contributes little to total HDR distortion.
  3. Quantifying the correlation between the Spatial Detail signal of processed images and corresponding original images provides a systematic method to choose the best combinations of HEVC-encoded resolutions and bitrates to minimize overall HDR distortion.

Our key objectives in this presentation is to provide a practical method that can help ensure great HDR experiences at any resolution.

By clicking the "Download Paper" button, you are agreeing to our terms and conditions.