Joint view filtering for multiview depth map sequences
Ri Li, Dmytro Rusanovskyy, Miska M. Hannuksela, Houqiang Li · 2012
Multi-view depth data representing the real-world 3D scenery can be inconsistent in the inter-view direction. Such inconsistency of depth data can distort the performance of the inter-view motion-compensated prediction (MCP), if the Multiview Video Coding extension of the Advanced Video Coding standard (H.264/MVC) is used for coding of multiview depth map data. In this paper we propose novel joint-view depth filtering (JVDF), which improves inter-view consistency of the multiview depth map data. Depth map images from all available views are warped to the same viewpoint, which results in multiple estimates of a real-world depth value. Estimates that belong to a confidence interval are fused together through a weighted average to produce a “noise-free” depth value estimate. In the provided simulation results the JVDF applied to 3-view depth map data significantly improves the performance of H.264/MVC coding, providing up to 13% of bitrate reduction.