4:2:0 Indistinguishable From 4:4:4?

Perceptual Effects of HDR Subsampling for Dolby Vision, Blu- Ray and BBC. Georg Schruff, Jan Hoydem, Jan Froehlich. July 2019

A study was conducted to determine the perceptual effect of subsampling in three different HDR standards: Dolby Vision, Blu-Ray and BBC. The results indicated that 4:2:0 subsampled HDR video sequences weren’t distinguishable from 4:4:4 sequences.

Chroma Subsampling

“For the detail of a scene the human visual system relies mostly on luma (Y’). The luminance information is largely found within the green spectrum of the light (500-565nm). In comparison, the blue and red wavelengths contain much more chrominance information. The chroma can be reduced without a noticeable loss of detail in the image. This process is called chroma subsampling.” Subsampling can result in luminance errors and hue shifts. The sequences used in the study were subsampled using the 4:2:0 scheme.

HDR Subsampling

For the Y’CbCr subsampling in HLG (BBC) and PQ (Blu-Ray), Rec.2020 Y’CbCr conversion was used. The Dolby Vision footage was transformed using ICtCp. For the testing, a 55″ LG OLED TV C9 was used, which the researchers said was unable to accept HLG, therefore the HLG footage was transformed to PQ.

Processing

Refer to the chart to see how each of the sequences was processed. When stills of the sequences were cropped and enlarged, it could be seen that Blu-ray subsampling produced dark artifacts, the BBC subsampling resulted in less dark artifacts at the color boundaries, and the Dolby Vision processing had a softening effect on the image and showed bright artifacts along the edges between the colors.

The Study

Surround luminance was maintained at 0.6 nits and participants (22 in all, for the most part, students from Stuttgart Media University) were seated at a distance of twice the height of the display. Participants were shown two different sequences, one entitled “carousel”, the other called ”smith”. The ”carousel” sequence, shot at a fair at night, was chosen for its extreme color gamuts. The ”smith” sequence of a blacksmith welding near a fire was expected to exhibit fewer differences, due to the fact that it didn’t contain colors close to the boundaries of the color gamut. In the first part of the study, participants were shown the three different standards (BBC, Dolby Vision and Blu-Ray) in comparison to the original reference, not knowing which was A or B. In the second part, the sample group was shown cropped-in still frames of the same sequences as in part one.

Conclusion

The study showed that even though subsampling artifacts can have a larger effect in HDR than in SDR, subsampled HDR footage wasn’t distinguished by the sample group when shown as a video sequence. Only when examining cropped-in still frames was the audience able to see differences. As the study sample consisted of only two sequences with no shots of natural subjects (daylight scenes with saturated colors, landscapes or human faces) shown on a display with a color gamut volume smaller than P3, it would be difficult to generalize from the results.