We present a subjective quality assessment of compressed 4K ultra-high-definition (UHD) videos in an immersive viewing environment. Three state-of-the-art video coding techniques, i.e., advanced video coding (AVC), high efficiency video coding (HEVC), and VP9. We aim at investigating added values of UHD over conventional high definition (HD) in terms of perceptual quality. The database includes video sequences and subjective ratings. Details of the data can be found in our paper.
Ten source videos (nine recorded sequences and one selected from the open source video) were considered in database. Source video sequences were captured using a RED EPIC-M digital cinematographic camera with the RED PRO ZOOM 18-85 mm lens. The recorded raw video sequences were saved in the equipment's own format, i.e., REDCODE (.r3d) raw data, which are virtually lossless. Recording parameters (e.g., ISO, shutter speed, focal length, etc.) were carefully adjusted depending on the scene. The original videos had the 5K resolution (5120x2700 pixels) at a frame rate of 30 fps, which were cropped to the 4K UHD resolution (3840x2160).
Total 240 distorted videos were generated from 10 reference videos. The resolutions of the stimuli used in our experiments are 4K UHD and 4K UHD up-scaled from FHD, where the latter assumes watching FHD videos at the 4K UHD resolution on a 4K UHD screen. To obtain reference FHD videos having the same contents with the reference 4K UHD videos, the 4K UHD sources were down-scaled to the FHD resolution. We employed three state-of-the-art video compression methods, AVC, HEVC, and VP9. Distorted video sequences that having various quality levels (low to high quality) and bitrates (low to high bitrates) based on extensive preliminary experiments. Four quality levels wre selected for each combinations of encoding methods and resolutions. The compressed videos having the FHD resolution were up-scaled to the 4K UHD resolution.
Subjective ratings were collected from twenty four subjects (eight females and sixteen males). Only part from start to 10 seconds was used in the experiment. The absolute category rating with hidden reference (ACR-HR) with 11 levels methodology was used. The excel table contains subjective scores.
Video sequences and subjective data may be used for research and development puposes only. Commercial applications are not allowed. Please cite following paper when you use this database:
For each download link, you can download zip files including eight distorted video sequences or encoded bitstream sequences. The videos are formatted as .yuv for all codecs. The bitstream sequences are formatted as .264, .bin, and .webm for AVC, HEVC, and VP9, respectively. These files are named as follows:
(source video)_(codec type)_(resolution type)_(distortion level)
S01_A_02_03.yuv, S04_A_01_04.264, S10_H_02_01.bin, etc.
For decoding bistreams (.264, .bin, and .webm), please refer this description and attached zip file.