AVS2-3D is the 3D extension of AVS2 (audio video coding standard 2), which is under the development of AVS Working Group. In current design of AVS2-3D, the picture-level global disparity vector (GDV) is adopted as the disparity vector (DV) for all blocks in a picture. However, due to the objects with varying depth within one picture, the accuracy of GDV is not enough. In this paper, a block-level depth information based DV derivation method is proposed for AVS2-3D. In order to get a more accurate DV for current block, two corresponding depth blocks are used in this method. Compared with the original GDV in AVS2-3D, it achieves an average of 3.43% and 2.09% bit-saving for all coded three views and synthesized six views, respectively. Additionally, the proposed design is also compared with the depth oriented neighbouring block based DV derivation (DoNBDV) of 3D-HEVC in AVS2 platform and the experimental results show that the method proposed provides an average of 0.35% and 0.36% coding gain compared with the DoNBDV.
[1]
Wen Gao,et al.
AVS2 ? Making Video Coding Smarter [Standards in a Nutshell]
,
2015,
IEEE Signal Processing Magazine.
[2]
G. Bjontegaard,et al.
Calculation of Average PSNR Differences between RD-curves
,
2001
.
[3]
Wen Gao,et al.
AVS2—Making Video Coding Smarter
,
2015
.
[4]
Bernhard P. Wrobel,et al.
Multiple View Geometry in Computer Vision
,
2001
.
[5]
Andrew Zisserman,et al.
Multiple View Geometry in Computer Vision (2nd ed)
,
2003
.