Edge and motion-adaptive median filtering for multi-view depth map enhancement

  • Authors:
  • Erhan Ekmekcioglu;Vladan Velisavljevic;Stewart T. Worrall

  • Affiliations:
  • Centre for Comm. Systems Research, University of Surrey, Guildford, UK;Deutsche Telekom Laboratories, Technische Universität Berlin, Germany;Centre for Comm. Systems Research, University of Surrey, Guildford, UK

  • Venue:
  • PCS'09 Proceedings of the 27th conference on Picture Coding Symposium
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a novel multi-view depth map enhancement method deployed as a post-processing of initially estimated depth maps, which are incoherent in the temporal and inter-view dimensions. The proposed method is based on edge and motion-adaptive median filtering and allows for an improved quality of virtual view synthesis. To enforce the spatial, temporal and inter-view coherence in the multiview depth maps, the median filtering is applied to 4-dimensional windows that consist of the spatially neighbor depth map values taken at different viewpoints and time instants. These windows have locally adaptive shapes in a presence of edges or motion to preserve sharpness and realistic rendering. We show that our enhancement method leads to a reduction of a coding bit-rate required for representation of the depth maps and also to a gain in the quality of synthesized views at an arbitrary virtual viewpoint. At the same time, the method carries a low additional computational complexity.