linkedin twitter facebook email


A novel light field representation and compression approach for VR

Virtual reality no longer belongs to the realm of the far-distant future. Using light-field imaging, it is possible to capture natural scenes in a way that allows them to be leveraged in a virtual reality environment afterwards. Unfortunately, light-field imaging yields extremely large amounts of data.

In this demo, researchers from imec - Ghent University (in collaboration with TU Berlin) present their experimental and novel ‘Steered Mixture-of-Experts’ (SMoE) technique – which makes processing, analyzing and compressing light fields more feasible. Their methodology differs drastically from the standard image processing paradigms, which have been struggling to scale towards new image types. Concretely:

  • The imec - Ghent University and TU Berlin approach does not code the light field 'pixels' captured by the camera, but instead reconstructs a model that could have produced these samples.
  • It then transforms this statistical model into bits and bytes while making sure the disk space requirements are kept at a minimum, yielding far beyond state-of-the-art compression ratios.
  • From this model, high-quality reconstructions of natural scenes can be instantly extracted, shown at any angle or position, at any desired resolution.
Copyright imec 2017