Accurate light and colour reproduction in high dynamic range video compression.

High Dynamic Range (HDR) imaging has the potential to replace traditional Low Dynamic Range (LDR) imaging due to HDR’s capability to accurately capture and reproduce the entire spectrum of visible lighting conditions with full colourimetric precision in any scene. However, this ability comes at the cost of significantly increased storage and transmission requirements compared to traditional LDR imaging. These costs together with additional challenges in capturing and delivering HDR video, for example ghosting artefacts, peak luminance of current HDR displays etc., are currently limiting the faster adoption of HDR imagery and the eventual replacement of traditional LDR imaging and video techniques. This thesis focuses on how to deliver high-fidelity HDR video with minimal storage/transmission requirements. To answer such a multi-faceted question, the thesis first provides an overview of HDR imaging and video pipeline followed by a detailed discussion on existing HDR video compression algorithms and quality assessment (QA) techniques for HDR image and video. This background information provides an in-depth review of the overall progress made to date and also highlights the current outstanding issues. The thesis subsequently assesses end-user preference of HDR video content over LDR video content using a rating- and a ranking-based psychophysical experiment. Results from this assessment suggest that there exists a statistically significant difference between the HDR representation of a scene and its LDR counterparts where given the option, the former is preferred by end-users as HDR provides a more realistic viewing experience. Having established the preference for HDR video, a comprehensive objective and subjective study is undertaken of a number of published/patented HDR video compression algorithms by means of several objective QA metrics and psychophysical studies. This resulted in an in-depth understanding of the advantages and shortcomings of existing solutions. Results obtained demonstrate that non-backward compatible compression algorithms are able to deliver high-fidelity HDR video at significantly lower storage/transmission costs compared to backward compatible algorithms. Also, perceptual QA metrics exhibit a high to very high correlation with subjective video quality assessment. Based on this in-depth understanding of the design requirements and philosophy of HDR video compression algorithms, this thesis proposes and evaluates a novel HDR video compression algorithm. This new algorithm is shown to outperform existing state-of-the-art algorithms both in terms of image reconstruction quality and transmission requirements