No comments

ABSTRACT

Until now, most of the research in the field of high dynamic range (HDR) video has centred on the use of non-real-time graded images which have been adjusted to look correct on a known reference screen in a reference environment.

For live television, without the luxury of grading, it is important that images captured directly by the camera look correct. So the HDR system’s end-to-end opto-optic transfer function (OOTF), which maps the light captured at the camera sensor to the light output from the display, is of paramount importance.

Furthermore, it is critical that the artistic intent of the video is preserved when rendered for the viewer with a different screen in a different viewing environment.

The authors present results of two subjective tests. The first test determines the most suitable OOTF for a reference environment and display; the second test determines how this transfer function could be adjusted so the high dynamic range video signal can be displayed on a range of different brightness displays whilst maintaining artistic intent.

INTRODUCTION

High Dynamic Range video (HDR) is a relatively new technique which allows the content producer to more accurately reproduce an image without the suppression of highlights usually associated with conventional video. Experiments show that there is a preference for high dynamic range video displayed on a high brightness monitor over a conventional television displaying standard dynamic range (SDR) video content (Hanhart et. al. (1)).

The television viewing experience has traditionally been defined in terms of a reference screen (EBU Tech 3320(2), ITU-R BT.1886(3))) being viewed in a reference environment (Teear (4)). At the time of writing HDR displays range in peak brightness from approximately 500 cd/m2 to approximately 4000 cd/m2. The authors expect brightness levels to increase as technology matures and new technologies reach the market. It is therefore critical that an end-to-end transfer function is chosen that allows artistic intent to be maintained on screens of differing peak brightness.

In this document we first present results of experiments undertaken to find the most suitable reference OOTF. The reference OOTF is proposed for a reference HDR monitor, under reference viewing conditions. In the second set of experiments we explore this reference OOTF further, to determine how it can be adjusted to display a high dynamic range video (Borer and Cotton (5)) signal on a range of different brightness screens under reference background lighting conditions.

In these experiments we use an end-to-end power function, or system gamma as proposed in Borer and Cotton (5), as the OOTF and adapt it for different brightness displays by changing the value of gamma, as illustrated in Figure 1. Our results show the change in system gamma that is required to obtain the best perceptual match between a signal graded on a given screen in a reference environment, and the same signal on any other screen.

Results of a supplementary test show that as the system gamma adjustments occur in the linear domain, they are independent of the exact choice of electro-optical or opto-electronic transfer functions.

BACKGROUND

The role of a reference OOTF is to map data from a scene captured by a camera to the display in a reference viewing environment. The reference OOTF should ideally be independent of any artistic adjustments so that, with adjustment to only the OOTF, the artistic intent can be maintained on a range of non-reference displays and in a range of viewing conditions.

With legacy SDR, reference displays were similar to those TVs found in the home, predominately CRT displays, therefore adjustments to the OOTF were not deemed necessary. With HDR, and a wide range of display technologies and viewing brightnesses, the need for suitable adjustments to a reference OOTF is absolutely required to maintain the artistic intent of the video.

Without any artistic adjustments or grading, such as in a live recording, the role of the OOTF is therefore to make the displayed images look as close as possible to the actual scene. Source linear scene-referred files were obtained from Fairchild (6) and from tests undertaken by Arnold & Richter (ARRI) and the Stuttgart Media University (Frölich et. al. (7)).

The source images were not graded, the colours and tones within them represent the actual scene. Therefore, in the first experiment we asked our viewers to rate the images on “naturalness”, or “realism”. That is, “if they were standing next to the camera, how natural do the images look?” Note that this is distinct from asking the viewers to choose their preferred image, which would include an element of artistic appreciation. For this comparison we were looking to achieve the most realistic end-to-end system.

The artistic choices to achieve the “best” image are a distinct and separate part of the television production process.