A New In-Loop Filter for Depth Map Coding in HEVC

Abstract

A depth-map is used to synthesize virtual texture views in the multi-view plus depth (MVD) format. In conventional video coding, a coded depth-map often suffers from compression artifacts along object boundaries, which have a negative effect on the quality of rendered images in the view synthesis process. To address this problem, we propose a depth-map boundary filtering technique to eliminate coding artifacts while preserving sharp edges. This can be mathematically formulated as a L0-norm minimization problem. This filtering process is cascaded with the de-blocking filter in the emerging HEVC video coding standard to result in a new in-loop filter. Experimental results are given to show that the subjective and objective quality of the synthesized views is enhanced by the introduction of the new in-loop fitler.