Bit-Rate Evaluation of Compressed HDR Using SL-HDR1

From a signal standpoint, what differentiates High Dynamic Range (HDR) content from Standard Dynamic Range (SDR) content is the mapping of the pixel samples to actual colors and light intensity. Video compression encoders and decoders (of any type) are agnostic to that ? the encoder will take a signal, compress it, and at the other side, the decoder will re-create something that is ?about the same? as the signal fed to the encoder. Existing encoders may have been optimized to provide good results with SDR, but HDR will still flow through, albeit possibly requiring higher bit rates for good reproduction.

Some of the HDR standards include dynamic metadata to help a display device render the content based on its capabilities. Some standards transmit a native HDR signal with metadata that allows the creation of anything from the original HDR all the way down to SDR. SL-HDR1, on the other hand, does the opposite: An SDR signal is transmitted, with metadata to inverse tone map to HDR. This metadata adds overhead to the video elementary stream.

This paper focuses on the required bit rates to produce a final HDR signal over a compressed link. We compare encoding SMPTE-2084 PQ HDR signals directly versus using SL-HDR1 to generate an SDR signal plus dynamic metadata. The comparison is done objectively by comparing the PSNR of the decoded signal. The SMPTE-2084 HDR signal is used as a reference at a fixed bit rate, and the bit rate of the SL-HDR1 encoded signal is varied until it matches the PSNR, over a range of source material. The evaluation is done for both AVC (H.264) and HEVC (H.265). This is similar to the work by Touze and Kerkhof published in 2017, but using commercial equipment.

Ciro Noronha | Cobalt Digital Inc. | Champaign, Illinois, USA
Kyle Wilken | Cobalt Digital Inc. | Champaign, Illinois, USA
Ryan Wallenberg | Cobalt Digital Inc. | Champaign, Illinois, USA