A Real Haze Video Database for Haze Level Evaluation
Ying Chu, Guoxing Luo, Fan Chen · 2021
Current video quality assessment methods mainly deal with videos with distortions generated in compression or capture process, few researchers focus on perceptual degradation caused by heavy haze in real outdoor scenes. However, assessment of haze level in such application scenario is important. For instance, in the advanced driver assistance system the accurate evaluation of haze level helps to properly release warning signals to drivers, which may avoid latent traffic accidents and save people's lives. It is necessary to develop studies on haze video quality assessment for real outdoor scenes, and to build benchmark dataset for haze videos. To fill the gap of this benchmark lack, a real haze video database called RHVD was designed and constructed. The dataset contains 403 haze videos captured in real outdoor scenes, which were collected from the Flickr video sharing platform and post-edited in our lab. We conducted subjective quality assessment experiments on the RHVD dataset. Forty non-experts were recruited and asked to rate the haze levels for each of the video in the dataset. The original ratings were screened and normalized to calculate the final subjective scores. The RHVD dataset is free to download from the following website: https://drive.google.com/file/d/1zY0LwJyNB8u1JTAJU2X7ZkiYXsBX7BF/view?usp=sharing, or https://pan.baidu.com/s/11p78zwMXmUHMGDj8WDanPA with password RHVD.