Abstract:
Wedgelet-based Depth Modeling Mode (DMM) in the 3D extension of HEVC (3D-HEVC) divides a depth block into two segments by a straight line, which can hardly approximate we...Show MoreMetadata
Abstract:
Wedgelet-based Depth Modeling Mode (DMM) in the 3D extension of HEVC (3D-HEVC) divides a depth block into two segments by a straight line, which can hardly approximate well complex edges in depth blocks, and consequently deteriorates coding performance of the depth blocks. In order to improve the depth coding performance based on better block representation, two straight lines or two parabola curves are jointly used to separate the block in the Wedgelet-based DMM. While two lines or curves can achieve significantly better approximation to complex edges, it introduces a remarkably increasing number of wedgelet templates, which in turn greatly exacerbates coding complexity and compromises coding performance. To attack the problem, uniform down sampling of the templates is applied coupled with template merging based on similarity and symmetricity. The proposed method is evaluated on the platform of the 3D-HEVC reference software HTM-15.2, which shows an average of 0.24% coding gain in terms of BD-rate under the All-Intra configuration compared to the Wedgelet-based DMM in the 3D- HEVC.
Published in: 2017 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)
Date of Conference: 12-15 December 2017
Date Added to IEEE Xplore: 08 February 2018
ISBN Information: