3D Video Disparity Vector Derivation Using Depth Subsets

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current 3D video coding methods face inefficiencies in disparity vector derivation from depth data, leading to high computational complexity and storage requirements, especially when dealing with large depth blocks and potential parsing errors in motion vector prediction.

Innovation Solution

Deriving disparity vectors from subsets of depth block samples, such as four corner samples, and generating additional motion vector candidates by reducing precision or adding offsets to existing candidates, reducing data access and computations while maintaining coding performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If disparity vectors are derived from all depth samples in a depth block, then measurement precision is improved, but computational complexity increases

Engineering Contradiction:
Improvedisparity vector derivation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the depth block into multiple regions (e.g., left region and right region) and derives disparity vectors separately for each region. This segmentation allows the system to process fewer samples per region while maintaining overall accuracy, thus reducing computational complexity without sacrificing measurement precision.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts only the necessary depth samples (e.g., boundary samples or region-specific samples) needed for disparity vector derivation, rather than processing all depth samples. This extraction approach reduces the number of computations required while preserving the essential information needed for accurate disparity calculation.

Inventive Principle:
Principle #2Taking out (Extraction)

2Productivity

If more motion vector candidates are generated, then coding performance is improved, but bitrate increases

Engineering Contradiction:
Improvecoding efficiencyVSAvoidbitrate
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent applies different precision levels to different motion vector candidates. Instead of maintaining full precision for all candidates, it uses reduced precision for certain candidates (e.g., derived candidates) while maintaining full precision for others (e.g., original candidates). This local quality differentiation improves coding efficiency by providing diverse candidates while controlling bitrate through selective precision reduction.

Inventive Principle:
Principle #3Local quality

3Measurement precision

If depth data is stored and processed in full resolution, then measurement precision is improved, but storage space increases

Engineering Contradiction:
Improvedepth data accuracyVSAvoidstorage space
Core Design Contradiction:
Measurement precisionVSVolume of stationary object

Solution Approach 1:

The patent applies partial action by processing only a subset of depth samples (e.g., boundary samples or region-specific samples) rather than all depth samples. This approach maintains sufficient measurement precision for disparity vector derivation while significantly reducing the storage space required for depth data, as the system only needs to store and process the essential subset of samples.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2727366B1Method and apparatus of motion and disparity vector derivation for 3D video coding and hevc
Publication Date: 2018.10.03 MEDIATEK INC
  • EP2727366B1 patent drawingFigure 1A~1B
  • EP2727366B1 patent drawingFigure 2~3
  • EP2727366B1 patent drawingFigure 4~5

AI summary

A method and apparatus for deriving MVP (motion vector predictor) for a block for three-dimensional video coding or multi-view video coding are disclosed. Embodiments according to the present invention replace an unavailable inter-view MV of one neighboring block with a disparity vector derived from depth data of a subset of a depth block corresponding to one neighboring block. A method and apparatus for generating additional candidates for motion vector prediction associated with Merge mode or AMVP (Inter) mode for a block are disclosed. Embodiments according to the present invention generate one or more additional MVP candidates to add to the MVP list if the MVP list size is less than a given list size. The additional MVP candidates are generated either by reducing precision of an available MVP in the MVP list or by adding an offset to the available MVP in the MVP list.