HDR2014 - A high dynamic range image quality database

Abstract

High dynamic range (HDR) imaging has attracted a lot of attention and enthusiasm in the last decades. With the quick advances of sensor technologies, even consumer level digital cameras are capable of capturing HDR images. However, a vast majority of nowadays displays still only have 8-bit color depth, and this leads to the widely studied topic of displaying HDR images on low dynamic range (LDR) devices or tone mapping. On the other hand, the recent emergence of 10-bit display devices brings the possibility of direct visualization of HDR images. So a natural question to ask is whether existing popular image quality metrics designed for and validated on LDR (8-bit) images perform equally well for HDR (10-bit) images. In this paper we propose a new and dedicated High Dynamic Range image quality database (HDR2014). That database is composed of 192 images with four kinds of distortions applied on 6 reference images. More specifically, we use 8 distortion levels for the artifacts of JPEG/JPEG2000 compression, white noise injection and Gaussian blurring. Twenty-five inexperienced viewers were involved in the subjective viewing test. Images were displayed on a pair of carefully calibrated 8-bit LDR and 10-bit HDR monitors and the subjective scores on both of which were recorded. We then tested some ubiquitous and state-of-the-art IQA metrics on the HDR2014 database. Experimental results show that HDR monitor indeed improved perceptual quality of the visual stimuli, as compared to LDR ones. And several existing IQA metrics are still doing well on HDR images, yet performance of some metrics drop significantly.