Volumetric Reconstruction with Compressed Data

Volumetric reconstruction is one of the 3D processing technologies for multi-view systems. The existing methods of volume reconstruction have been proposed by exploiting the original views with many constraints. These algorithms may not be suitable for the distributed camera system where multiple views are transmitted via lossy networks before processing at the computing centre. Therefore, this paper proposes a novel volumetric reconstruction from the compressed multi-view images. The algorithm starts with the depth registration, and then the initial volume is refined. Finally the colour selection scheme provides the realistic colour to the volume thereby achieving subjective quality of rendering views as shown in the subjective results. The performance of the proposed scheme reconstructing with the compressed views may be inferior than reconstructing with the original views up to 3 dB, but it can be superior over 3 dB at low bitrate.