Depth Image-Based Rendering for Full Parallax Displays :
Rendering, Compression, and Interpolation of Content for Autostereoscopic Poster and Video Displays
- Zugl.: Kiel, Univ., Diss. 2014
Advancements in production and display techniques allowed for novel displays to emerge that project a high-resolution light field for static poster content and video content, as well. These displays allow a full parallax, hence an audience can perceive a stereoscopic view of a scene without special glasses, which adjusts to the observer’s position. The application of such displays are public places where the audience does not wear special glasses and is not restricted in movement.
The rendering, storage, and transfer of the large amount of data required by those displays is a challenge. The image data for a static poster display is about 200 GB and the data rate for video displays are to be expected two to four orders of magnitude higher than HDTV.
In this work the challenges are met by utilising Depth Image-Based Rendering to reduce the amount of data at the very beginning, during rendering. A fraction of the full amount of colour and depth images are rendered and used to interpolate the full data set. The rendering with state of the art ray tracers is described and a novel method to render image data for full parallax displays using OpenGL is contributed, that addresses some shortcomings of previous approaches.
For static poster displays a scene based representation for image interpolation is introduced, which efficiently utilises multi-core processors and graphics hardware for parallelization, found on modern workstations.
The introduced approach implements lossy compression of the input data, and handles arbitrary scenes, using a novel Best-Next-View selection algorithm. For video displays the real-time constraint does not allow for a costly interpolation or scene analysis. Hence, a novel approach is presented that uses a basic and computational inexpensive interpolation, and combines the interpolation results of different image representations without introducing prominent artefacts.