Improving the coding performance of 3D video using guided depth filtering

Autostereoscopic displays visualize a 3D scene based on encoded texture and depth information, which often lack the quality of good depth maps. Therefore, display manufacturers introduce different filtering techniques to improve the subjective quality of the reconstructed 3D image. This paper investigates the coding performance when applying depth filtering in a pre-processing step. As an example, guided depth filtering is used at the encoder side, which results in a 1.7% coding gain for 3D-HEVC and 8.0% for Multiview HEVC. However, applying additional filtering at the decoder side might deteriorate the subjective quality. Therefore, adaptively filtering based on the applied pre-processor filter is suggested, which can be done using a supplemental enhancement information message. For natural content, a gain of 5.7% and 9.3% is reported using this approach.