Rendering multi-view plus depth data on light-field displays

Abstract

This paper presents an approach for rendering heavily extrapolated novel views to be used as input for light-field displays. This view generation method builds on a combination and enhancement of existing methods. The interpolation quality is assured by detecting and keeping the most reliable gap area information from the content using depth layers. Concerning the extrapolation process, which is the most important part of this paper, we implemented an algorithm that prefers isophotes lines in order to reconstruct objects and patterns using gradient filling and Poisson reconstruction. Using the algorithms described, it is possible to generate wide baseline light field data from Multi-View plus Depth (MVD) data of moderate baseline. The approach is demonstrated by generating interpolated and extrapolated views for feeding a HoloVizio large-scale display with captured video data.