A Biologically Inspired Approach to Making HDR Video Quality Assessment Easier

摘要

Quality is easy to recognize by the eye. Quantifying it can be more problematic. For high dynamic range (HDR), it is even harder because convenient methods, such as peak signal-to-noise ratio and structural sim- ilarity metric, do not work well. Some newer HDR-aware metrics are now available; how- ever, they can be slow and computationally cumbersome, therefore discouraging use in realtime monitoring. This paper introduces an alternative, simpler technique that could be used in post-production and realtime moni- toring during distribution. Specifically, this paper presents a biologically inspired method of creating a "spatial detail" representa- tion that can be used instead of the original HDR image as a basis for quantifying video distortions and artifacts. Simple correlation analysis of the luma spatial detail signal yields an intuitive and sensitive metric of dis- tortion in HDR video. As an added benefit, the spatial detail signal provides a guide that enables distortions in HDR highlights to be evaluated separately from the remain- der of the video.