Subjective Quality Evaluation of High Dynamic Range Video and Display for Future TV by Philippe Hanhart, Touradj Ebrahimi, Pavel Korshunov and Yvonne Thomas. SMPTE Motion Imaging Journal, May 2015
The following excerpts are from a study carried out at the European Broadcasting Union and École Polytechnique Fédérale de Lausanne in 2015.
Why 100, 400, 1000 and 4000 nit levels were chosen. How clips were mastered and tone-mapped
The video material was specifically tailored for display at 100, 400, 1000, and 4000 nits. These four dynamic range levels were selected to be representative of key use cases, as opposed to being uniform perceptual distances. The maximum luminance of 100 nits corresponds to standardized reference monitor levels used for studio color grading and broadcast calibration, 400 nits correspond luminance levels in current high-quality liquid crystal display (LCD) and light-emitting diode backlit LCD TVs, 1000 nits correspond to the highest levels found today in consumer TVs, and 4000 nits corresponds to the max luminance level of the HDR display used in this experiment. The displayed video was constructed as follows:
1)4000 nit version: manually graded by professional colorists from the original video, captured using one of the highest dynamic range digital video cinema cameras (Arri Alexa with 14 f-stop range)
2) 1000 nit version: tone-mapped from manually graded 4000 nit version
3)400 nit version: tone-mapped from manually graded 4000 nit version
4)100 nit version: tone-mapped from manually graded 4000 nit version.
For the tone mapping, an automated proprietary tone mapping algorithm was used. This algorithm was designed to preserve overall appearance to the input (graded) version. It was not intended to perform enhancements or bias importance to specific image regions s often occurs in human-guided color grading). The input was 12 bits per color in a domain that has characteristics of gamma and log nonlinearities, as suited for HDR.³ The video sequences were uncompressed. The combination of high bit depth and uncompressed video is intended to remove secondary issues of dynamic range effects on needed bit depth and compression algorithm parameters because the study’s aim was to isolate the question of range, start with maximum luminance.
Testing Methodology, Surround Illumination
Evaluations carried out at the EBU relied on expert viewers. A time sequential presentation with both a forced-choice preference, as well as a horizontal preference scale, was conducted, using a 4000 nit graded content as a hidden reference. The hidden reference was shown in every paired comparison, with a randomized position. This methodology was derived from a well-known video comparison methodology The forced-choice preference is a binary scale that directly identifies which condition is preferred, whereas the horizontal preference scale, which is a continuous version of the comparison scale,ª provides a finer comparison of the two conditions. Two viewing distances were tested: 3 H (~1.5 m), a recommended distance for subjective tests, and 2.7 m, a typical distance from TV set in a home environment. The display size was 42 in.
In the EBU tests, the illumination surrounding the display was set to 10 lux for most test groups, and then to 24 lux for one test group to test the effects of ambient illumination levels affecting viewers’ light adaptation as well as screen reflectivity consequences. In the EPFL tests, the illumination surrounding the display was set to 20 lux and was thus in the same range as in the EBU tests’ backlight settings. [grading suites are 16 lux]
The scores in the EBU and EPFL tests included both a forced choice and horizontal preference scales, as shown in Fig. 2. In the EBU tests “Left” was replaced with “A” and “Right” with “B” be cause of the time sequential display mode used.
Subjects were asked to rate the overall quality of pairs of displayed video sequences. To select a score, subjects were instructed to consider characteristics such as color rendition, quality of the reproduction of skin tones, details of shadows in the scene, contrast and the details of highlights, and presentation of light sources appearing in the scene.
Results
Figure 4 shows the overall results obtained at the EBU and the EPFL tests for forced-choice scores, with their respective confidence intervals. It can be seen that the forced-choice preferences increase with the increase in the peak brightness from 100 to 4000 nits, to reach 0.5, which corresponds to the preference of 4000 nit sequences (i.e., the hidden reference), which should be theoretically random (50%), since both of the paired comparison video sequences were identical. These results show that there is a significant preference toward 4000 nit displayed content when compared to other alternatives considered in the tests. The relatively flat region from 100 to 1000 nits as compared to the steep rise for 4000 nits indicates the 4000 nit version was substantially different and preferred. It does not necessarily mean that the 100, 400, and 1000 nit versions were close together in appearance and preference. That question requires the second portion of the study, the preference scale responses.
Hanhart, Philippe & Korshunov, Pavel & Ebrahimi, Touradj & Thomas, Yvonne & Hoffmann, Hans. (2015). Subjective Quality Evaluation of High Dynamic Range Video and Display for Future TV. SMPTE Motion Imaging Journal. 124. 1-6. 10.5594/j18555.
Previous studies by Dolby Laboratories showed that 1,000 nits satisfied only around 20% of participants.
The Daejeon Chronicles 
Leave a Reply