Decoder Motion Vector Derivation for Video Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing decoder-side motion vector derivation methods for video coding, such as bilateral matching and template matching, incur high computational load and system memory bandwidth due to the need to access various motion vector candidates and reference data, which hampers coding efficiency and increases bitrate.

Innovation Solution

The proposed method involves signaling an initial motion vector index, deriving a selected motion vector using bilateral or template matching, and refining it to determine a final motion vector predictor, while reducing the number of starting motion vectors and optimizing the search process to minimize computational load and memory access.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If decoder-side motion vector derivation methods (bilateral matching and template matching) are used to improve coding efficiency, then motion information accuracy is improved, but computational load and system memory bandwidth increase

Engineering Contradiction:
Improvemotion information accuracyVSAvoidcomputational load
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent applies preliminary action by generating a reduced set of candidate motion vectors before the matching process. The encoder prepares and signals a limited number of candidate MVs to the decoder, which then performs bilateral matching or template matching only on these pre-selected candidates rather than searching the entire motion vector space. This preliminary preparation significantly reduces the computational load while maintaining motion estimation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent segments the motion vector search space by dividing it into multiple candidate vectors that are separately evaluated. Instead of performing exhaustive search on all possible motion vectors, the method segments the search into evaluating a limited set of candidate MVs (e.g., 2-4 candidates), performing matching operations only on these segmented candidates, thereby reducing overall computational complexity.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If decoder-side motion vector derivation methods are used to improve coding efficiency, then motion information accuracy is improved, but system memory bandwidth increases

Engineering Contradiction:
Improvemotion information accuracyVSAvoidmemory bandwidth
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The encoder performs preliminary action by pre-calculating and signaling a reduced set of candidate motion vectors to the decoder. This preliminary preparation eliminates the need for the decoder to access and process large numbers of reference picture blocks and motion vectors during the matching process, thereby significantly reducing memory bandwidth requirements while maintaining accurate motion estimation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent extracts only the essential candidate motion vectors from the full motion vector space and transmits them to the decoder. By taking out only the necessary candidates (a small subset of 2-4 MVs) rather than transmitting or processing all possible motion vectors, the method reduces the quantity of data that needs to be stored in and accessed from memory, thereby reducing memory bandwidth consumption.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If a full set of motion vector candidates is evaluated to ensure accuracy, then motion estimation precision is improved, but device complexity increases

Engineering Contradiction:
Improvemotion estimation precisionVSAvoidalgorithm complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The encoder performs preliminary action by pre-selecting and signaling a limited set of candidate motion vectors that are most likely to be optimal. This preliminary selection simplifies the decoder's task from evaluating all possible motion vectors to evaluating only the pre-selected candidates, thereby reducing algorithmic complexity while maintaining motion estimation precision through careful candidate selection.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent applies partial action by evaluating only a partial set of motion vector candidates (2-4 candidates) rather than the complete set of all possible motion vectors. This partial evaluation is sufficient to achieve accurate motion estimation because the candidate set is carefully constructed to include the most promising vectors, thereby reducing device complexity without sacrificing precision.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3342167B1Method and apparatus of decoder side motion derivation for video coding
Publication Date: 2020.12.30 MEDIATEK INC
  • EP3342167B1 patent drawingFigure 1~2
  • EP3342167B1 patent drawingFigure 3A~3B
  • EP3342167B1 patent drawingFigure 4

AI summary

Method and apparatus of video coding using decoder derived motion information based on bilateral matching or template matching are disclosed. According to one method, an initial motion vector (MV) index is signalled in a video bitstream at an encoder side or determined from the video bitstream at a decoder side. A selected MV is then derived using bilateral matching, template matching or both to refine an initial MV associated with the initial MV index. In another method, when both MVs for list 0 and list 1 exist in template matching, the smallest-cost MV between the two MVs may be used for uni-prediction template matching if the cost is lower than the bi-prediction template matching. According to yet another method, the refinement of the MV search is dependent on the block size. According to yet another method, merge candidate MV pair is always used for bilateral matching or template matching.