WO2015051533A1

WO2015051533A1 - Improved advanced residual prediction in 3dvc

Info

Publication number: WO2015051533A1
Application number: PCT/CN2013/085049
Authority: WO
Inventors: Kai Zhang; Jicheng An; Jian-Liang Lin
Original assignee: Mediatek Singapore Pte. Ltd.
Priority date: 2013-10-11
Filing date: 2013-10-11
Publication date: 2015-04-16

Abstract

This contribution proposes to move syntax elements related to 3D-HEVC from slice segment header to slice segment header extension. In this way, slice segment header in 3D-HEVC can be compatible with that in HEVC.

Description

IMPROVED ADVANCED RESIDUAL PREDICTION IN

3DVC

TECHNICAL FIELD

[0001] The invention relates generally to Three-Dimensional (3D) video processing. In particular, the presented invention relates to advanced residual prediction.

BACKGROUND

[0002] In the current 3D-HEVC, advanced residual prediction (ARP) was adopted to replace the simple inter-view residual prediction for its better coding efficiency. In ARP, the base view is considered as a mirror of the current view. A mirror block is located with an estimated disparity vector (DV) and a mirror reference block for the mirror block is located using the same motion vector (MV) as the current block. Pseudo residues are generated by subtracting the reconstructed samples in the mirror block and in the mirror reference block. Then pseudo residues are used to predict the real residues of the current block. Fig. 1 depicts a sketch of ARP. In ARP, a coding unit (CU) can decide to use the pseudo residual prediction or not. A flag is transmitted to signal the on/off control including the weighting factor for the pseudo residual prediction (ARP). The weighting factor could be 0 (ARP disabled), 1 (ARP enabled) or ½ (The ARP is enabled but the pseudo residues will be divided by 2). For each CU, the encoder determines to enable or disable the ARP according to the rate- distortion optimization (RDO) criterion, and signals the control flag in the bit-stream.

[0003] It plays a critical role to locate the mirror block in the base-view accurately since pseudo residues are similar to the real residues only if they are generated by similar samples. In ARP, the estimated DV is derived by a neighbouring block disparity vector (NBDV) procedure. The first encountered DV from a temporal or spatial neighbouring block is utilized by ARP as the estimated DV for the current block. Since such an estimated DV is not accurate enough, pseudo residual prediction may be inefficient in some cases.

SUMMARY

[0004] In light of the previously described problems, an adaptive disparity vector derivation (ADVD) method for advanced residual prediction (ARP) and an advanced temporal residual prediction (ATRP) method are proposed. In ADVD, three DV candidates can be chosen to provide more accurate pseudo residues. In ATRP, the concept of ARP is also applied to the disparity compensated prediction. In the combination of ADVD and ATRP, three MV candidates can be chosen in ATRP.

[0005] Other aspects and features of the invention will become apparent to those with ordinary skill in the art upon review of the following descriptions of specific embodiments.

BRIEF DESCRIPTION OF DRAWINGS

The invention can be more fully understood by reading the subsequent detailed description and examples with references made to the accompanying drawings, wherein:

[0006] Fig. 1 is a diagram illustrating a sketch of ARP, where PR represents pseudo residue;

[0007] Fig. 2 is a diagram illustrating the order to fill the DV candidate list;

[0008] Fig. 3 is a diagram illustrating the aligned temporal DV and the temporal DV;

[0009] Fig. 4 is a diagram illustrating the concept of the proposed ATRP;

[0010] Fig. 5 is a diagram illustrating the derivation of DMV;

[0011] Fig. 6 is a diagram illustrating the order to fill the MV candidate list in

ATRP;

[0012] Fig. 7 is a diagram illustrating exemplary reference blocks.

DETAILED DESCRIPTION

[0013] The following description is of the best-contemplated mode of carrying out the invention. This description is made for the purpose of illustrating the general principles of the invention and should not be taken in a limiting sense. The scope of the invention is best determined by reference to the appended claims.

[0014] An adaptive disparity vector derivation (ADVD) method is proposed to improve the ARP coding efficiency. In ADVD, three DV candidates are derived from temporal and spatial neighbouring blocks. The candidate list is filled according to the order depicted in Fig. 2. The aligned temporal DV is obtained from the aligned block, which is located by a scaled MV to the collocated picture, as depicted in Fig. 3. The aligned block can be obtained as follows. The left-top coordinate of the aligned block is equal to the left-top coordinate of a reference block inside or outside of the current block, added by the MV of the current block scaled to the collocated picture. Fig. 7 demonstrates exemplary reference blocks. A new DV candidate is appended into the list only if it is not equal to any DV candidate already in the list. An encoder can determine the best DV candidate used in ARP according to the RDO criterion, and signal the index to the decoder.

[0015] An advanced temporal residual prediction (ATRP) method is proposed for the disparity compensated prediction. In ATRP, the disparity parameters of the current PU are applied to the corresponding block in a temporal reference picture in the same view to generate the reference residual in temporal direction. The corresponding block in the temporal reference picture is located by a derived motion vector (DMV), which is the motion vector (MV) of the reference block which the current DV is pointing to in the reference view. The reference block can be obtained as follows. The left-top coordinate of the reference block is equal to the left-top coordinate of a reference block inside or outside of the current block, added by the DV of the current block. Fig. 7 demonstrates exemplary reference blocks. The concept of proposed ATRP is illustrated in Fig. 4, and the derivation of DMV is illustrated in Fig. 5.

[0016] When ADVD and ATRP are combined, three MV candidates are derived for ATRP. The order to fill the three MV candidates is depicted in Fig. 6. The spatial MVs and temporal MVs are derived from the spatial and temporal merging candidate list and the first two available spatial and/or temporal MVs will be included into the MV candidate list after the DMV. An encoder can determine the best MV candidate used in ATRP according to RDO criterion, and signal the index to the decoder, similar to what is done in ADVD for ARP.

Claims

1. A method of generating a derived disparity vector (DV), wherein the derived DV of a current block is selected from multiple DV candidates.

2. The method as claimed in claim 1, wherein a predefined order is used to select and include multiple DV candidates in a DV candidate list.

3. The method as claimed in claim 2, the derived DV is selected from the DV candidate list.

4. The method as claimed in claim 2, wherein an encoder selects a best DV candidate from the DV candidate list as the derived DV for a current block according to a criterion.

5. The method as claimed in claim 4, a rate-distortion optimization criterion is applied to select the best DV candidate in a given DV candidate list.

6. The method as claimed in claim 2, wherein the derived DV is signaled explicitly to a decoder or derived implicitly by the decoder.

7. The method as claimed in claim 1, wherein the derived DV is used in procedures requiring a derived DV, including but not limited to advanced residual prediction (ARP), inter-view motion prediction, and inter-view motion/disparity merging scheme.

8. The method as claimed in claim 1, wherein an aligned temporal DV is used as a DV candidate, the aligned temporal DV is a DV of an aligned block in a collocated picture, and the aligned block is located using a motion vector (MV) of the current block scaled to a collocated picture.

9. The method as claimed in claim 8, wherein there are two collocated pictures at most, which are signaled to a decoder explicitly or derived by the decoder implicitly at slice level.

10. The method as claimed in claim 8, wherein there are two collocated pictures, the first one is signaled to a decoder explicitly and the second one is derived implicitly.

11. The method as claimed in claim 8, wherein the aligned block is obtained as follows: a left-top coordinate of the aligned block is equal to a left-top coordinate of a reference block inside or outside of the current block, added by a MV of the current block scaled to the collocated picture.

12. The method as claimed in claim 8, wherein the aligned temporal DV is a DV in an aligned block for reference list 0 or for reference list 1.

13. The method as claimed in claim 8, wherein there is more than one aligned temporal DVs used as the DV candidates.

14. The method as claimed in claim 8, wherein N aligned temporal DVs are treated as first N DV candidates in the candidate list, and N is equal to or large than 1.

15. A method of generating a derived motion vector (MV), wherein the derived

MV of a current block is chosen from several MV candidates.

16. The method as claimed in claim 15, wherein a predefined order is used to select and include multiple MV candidates in a MV candidate list, and the derived MV is selected from the MV candidate list.

17. The method as claimed in claim 16, wherein an encoder selects a best MV candidate from the MV candidate list as the derived MV for the current block according to a criterion, and the criterion includes rate-distortion optimization.

18. The method as claimed in claim 15, wherein a best MV candidate is signaled explicitly to a decoder or derived implicitly by the decoder.

19. The method as claimed in claim 15, wherein the derived MV is used in procedures requiring a derived MV, including but not limited to advanced temporal residual prediction (ATRP).

20. The method as claimed in claim 15, wherein the candidate MV comprises one or more MV in a reference block, for reference list 0 or for reference list 1 , and the reference block is located with the DV of the current block to an inter-view picture.

21. The method as claimed in claim 15, wherein N derived MVs are treated as first N MV candidates in a candidate list, and N is equal to or large than 1.

22. The method as claimed in claim 15, wherein the MV in a neighboring block of the current block is a candidate MV for the current block, the neighboring block is a temporal neighboring block or a spatial neighboring block, and a zero-MV=(0,0) is included as a MV candidate.

23. The method as claimed in claim 15, wherein the number of MV candidates denoted as N is greater than 1 , N is transmitted in a bit-stream by explicitly signaled in a sequence, view, picture, slice, CU or PU level, or N is implicitly derived.

24. The method as claimed in claim 23, wherein N is the same for all blocks in the same picture.