The following paper have been accepted for presentation at ICIP 2016 in Phoenix, Arizona, US.
Authors: X. Su, T. Maugey, C. Guillemot
Title: Graph-based Representation for Multiview Images With Complex Camera Configurations
Abstract: Graph-Based Representation (GBR) has recently been proposed for rectified multiview dataset. The core idea of GBR is to use graphs for describing the color and geometry information of a multiview dataset. The color information is represented by the vertices of the graph while the scene geometry is represented by the edges of the graph. In this paper, we generalize the GBR to multi-view images with complex camera configurations. Compared with previous work, the GBR representation introduced in this paper can handle not only horizontal displacements of the cameras but also forward/backward displacements, rotations etc. In order to have a sparse (i.e., easy to code) graph structure, we further propose to use a distortion metric to select the most meaningful connections. For the graph transmission, each selected connection is then replaced by a disparity-based quantity. Hence, the graph is constructed with just enough information for rendering the given predicted view. The experiments show that the proposed GBR achieves high reconstructing quality with less or comparable coding rate compared with traditional depth-based representations, that directly compress the depth signal without considering the rendering task.