Graph-Based vs Depth-Based Data Representation for Multiview Images

In this paper, we propose a representation and coding method for multiview images. As an alternative to depth-based schemes, we propose a representation that captures the geometry and the dependencies between pixels in different views in the form of connections in a graph. In our approach it is possible to perform compression of the geometry information and to preserve a direct control of the effect of geometry approximation on view reconstruction. This is not possible with classical depth-based representations. As a results, our method leads to more accurate view prediction, when compared to conventional lossy coding of depth maps operating at the same bit rate. We finally show in experiments that our representation adapts the amount of transmitted geometry to the complexity of the predictions that are performed at the decoder.