Back HDR models and methods for cinema postproduction

HDR models and methods for cinema postproduction

HDR models and methods for cinema postproduction

There is currently very substantial interest, both at academic and industrial levels, in the issues related to capture, processing and display of HDR images. HDR technology is widely seen in the industry as the next generation of video content because it is expected to deliver a better user experience through enhanced contrast, brightness and color. Given that at present the vast majority of display devices are LDR, at the same time that computer generated visual effects are often created as HDR data to be composited onto LDR real footage, we identify three main problems to address within this particular research objective:


  1. How to properly merge image sources with different dynamic range; in practice this is done via the software-assisted work of very skilled technicians/artists.

  2. Transformation of HDR images into LDR keeping detail visibility and natural appearance, a problem known as tone mapping (TM) in the computer graphics literature; conversely, generation of an HDR image from an LDR source (inverse TM). The literature on TM is abundant and many operators have been proposed: the best performing ones are based on perception and vision models, others work directly on the contrast or the gradient field, some allow for user interaction, but still TM for cinema remains a challenging open problem.

  3. Development of perceptual-based image quality metrics that assess the quality of the tone mapping. Current TM metrics are quite limited in that they deal only with luminance (not considering color) and are based on visibility thresholds, i.e. a very visible difference and a just noticeable difference are marked as errors of the same amount, and the global appearance of the image in the tone-mapped result is not considered by these metrics either; as a consequence, their results don't correlate well with those of psychophysical tests in which observers are asked to evaluate visual quality of TM outputs.


The aim of the research work in this proposal is to develop vision models and software methods allowing us to properly handle high dynamic range (HDR) content during cinema postproduction, with three main goals: combining images of different dynamic range, conversion of dynamic range (tone mapping (TM) and inverse TM), and evaluation of TM results.


The proposed line of research is multi-disciplinary, combining vision science (perceptual models from psychophysics, efficient coding from visual neuroscience), image processing (natural image statistics) and computer graphics.


To know more:







Principal researchers

Marcelo Bertalmío
Josep Blat


David Kane
Ricardo Marques