ABSTRACT

In the last few years, High Dynamic Range (HDR) has made major steps forward to become the next big broadcast technology. It is generally accepted that HDR will need a higher bit-rate because of its minimum quantisation and the fact that the images have much more details in the highlights and shadows.

However, it is insufficiently taken into account that most of these details will be preserved when performing an HDR downconversion using tone mapping. It remains unclear how the SDR (Standard Dynamic Range) bit-rate is influenced by HDR production.

Therefore, PSNR measurements are performed and a detailed explanation of the reasons is given. Moreover, it is not known for certain if the strong manipulation of the luminance component at down-conversion will produce artefacts like banding.

Information on the incoming bit-depth of the HDR with subsequent tone mapping is derived from a viewing test. 

INTRODUCTION

One of the main difficulties of introducing HDR is compatibility with the SDR-TV-sets already installed. To benefit from the higher dynamic range of modern TV cameras and to achieve better picture quality on legacy TV sets, tone mapping techniques such as the Sectional Tone Mapping introduced at IBC2016, can be used.

An intense manipulation of the luminance component in response to neighbouring pixels, enables a reproduction of a high scene contrast range on SDR displays. Compared to that of today’s SDR images, the down-converted pictures show much more detail in the shadows and highlights.

This leads to two important issues which will be considered in this paper:

First: An SDR picture after tone mapping, which was down-converted from HDR, could require a higher bit-rate to cope with the increased detail in the highlights and shadows. The bit-rate could even be higher than with PQ (Perceptual Quantiser) or HLG (Hybrid Log Gamma) signals because of a non-perfect bit-allocation at gamma.

This could not only be important for SDR-TV, but also for HDR-TV when using the base- and the enhancement-layer concept. In this case, the base-layer is designed to replace the existing SDR picture – but it is not the same for the aforementioned reasons. Due to the complexity of coding algorithms, it is difficult, if not impossible, to predict quantitative results.

When thinking about an interframe coding structure, the bit-rate of the I-frames could increase significantly. But at P- or B- frames, an opposite effect could be observed due to the higher homogenisation over time e.g. no changes in aperture. Therefore, motion-compensation can be more effective. In this paper, the results of objective and subjective tests of different test scenes and compression rates are given, that show which parameters in the picture influence the compressed bit-rate.

Second: The Perceptual Quantiser (PQ) is based on the Barten ramp [3] modelling the contrast sensitivity function of the human visual system (HVS) to get the best possible bitallocation and to avoid visible artefacts, like banding.

However, tone mapping adds another nonlinear function to the PQ curve so that the size of the steps and the OETF (Opto-electronic Transfer Function) changes can lead to undesirable effects, especially when encoded with 8 bits. The test results reveal the critical parts in the picture and the corresponding compression rate at which these artefacts are recognised by the viewers. 

Download the full technical paper below