Feedback-Assisted Adaptive Network Coded Cooperation for Wireless Networks

Abstract

Inaccurate depth map discontinuities are an important source of artifact and distortion in multiview systems and applications requiring depth-based view rendering such as 3DTV or free-viewpoint video. Such errors along depth discontinuities frequently occur during depth estimation and/or depth map coding. In this paper we present a novel method for correcting inaccuracies on the discontinuities along object borders within depth maps using color features available within the associated view. Erroneous depth map border areas are split into a set of fine, color-homogeneous regions and a region merging technique is used to form new depth map borders in conformance with the available color cues. The proposed algorithm can align depth discontinuities to object borders using only information within a single view. Results demonstrate the effectiveness of our method on multiview video-plus-depth test images.