Texture Super-Resolution in Multiview RGB-D Transmission

摘要

Auto-stereoscopic displays allow viewers to experience 3D content by displaying multiple camera views. To reduce bandwidth requirements the multiview video plus depth (MVD) representation can be used, where only a subset of these camera views needs to be transmitted together with the depth maps. The display can then apply depth-image-based rendering techniques to reconstruct the missing views. In this work we propose further reduction in bandwidth through downsampling of the texture views before encoding and then apply a dictionary-based super-resolution method during the upsampling process at the receiver. The depth videos are still transmitted at full resolution. Multiview high efficiency video coding (MV-HEVC) is used to separately encode the texture and depth videos because of their different resolutions. The visual quality of the decoded and reconstructed content was evaluated through objective and subjective testing giving fair to good and poor to fair quality results for texture downsampling by two and four respectively.