Consolidated Dataset and Metrics for High-Dynamic-Range Image Quality

Abstract

Increasing popularity of high-dynamic-range (HDR) image and video content brings the need for metrics that could predict the severity of image impairments as seen on displays of different brightness levels and dynamic range. Such metrics should be trained and validated on a sufficiently large subjective image quality dataset to ensure robust performance. As the existing HDR quality datasets are limited in size, we created a Unified Photometric Image Quality dataset (UPIQ) with over 4,000 images by realigning and merging existing HDR and standard-dynamic-range (SDR) datasets. The realigned quality scores share the same unified quality scale across all datasets. Such realignment was achieved by collecting additional cross-dataset quality comparisons and re-scaling data with a psychometric scaling method. Images in the proposed dataset are represented in absolute photometric and colorimetric units, corresponding to light emitted from a display. We use the new dataset to retrain existing HDR metrics and show that the dataset is sufficiently large for training deep architectures. We show the utility of the dataset on brightness aware image compression.

PU21: A novel perceptually uniform encoding for adapting existing quality metrics for HDR

“…To evaluate the new PU21 encoding and to choose the best variant from the four developed combinations, we employ the new PU21 encodings to predict HDR image quality in UPIQ dataset [22]. This dataset consists of over 3779 SDR and 380 HDR image pairs.…”

The effect of display brightness and viewing distance: a dataset for visually lossless image compression

“…They produce a visibility map with local information about the probability of detecting a difference between two images. Similar to compression algorithms, image quality and visibility metrics can be divided into two categories: hand-crafted [30,31,37], and trained using machine-learning methods [4,10,11,20,24,35,38]. Both quality and visibility metrics have been adapted to operate under different display luminance [1,19,20,35].…”

“…The predicted VLTs then can be used to automatically adjust compression parameters to achieve a trade-off between image visual quality and bit-rate. We evaluated both hand-crafted image quality metrics (PSNR, SSIM [30], MS-SSIM [29], FSIM [36], HDRVQM [22]) and one metric based on machine learning (deep photometric image quality metric: PU-PieApp) [20]. Additionally, we evaluate two visibility metrics: a high dynamic range visible difference predictor (HDRVDP3) [19,21], and a CNN-based deep photometric visibility metric (DPVM) [35].…”

Multi-Dimensional Feature Fusion Network for No-Reference Quality Assessment of In-the-Wild Videos

“…For NR-I/VQA, there are already some databases suitable for training deep learning networks, such as UPIQ [58], CVD2014 [59], KoNViD-1k [60], LIVE-Qualcomm [61], LIVE-VQC [62], etc. Network models trained for a specific database often perform poorly in terms of prediction in other databases due to the differences between individual databases.…”