Context-Based Motion Estimation Vector Refinement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The high computational complexity of motion estimation in video encoders, particularly in the H.265/MPEG-H HEVC standard, due to the need for sub-pel resolution refinement, poses challenges for real-time encoding and efficient coding efficiency, especially with the increased demand for High Definition (HD) and Ultra High Definition (UHD) content.

Innovation Solution

A context-based refinement method that determines a context from a set of lower-pel resolution vectors to sequentially test higher-pel resolution candidate vectors, reducing computational complexity while maintaining coding efficiency by selecting the motion estimation vector with the lowest coding cost.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If sub-pel resolution motion estimation refinement is performed using conventional interpolation-and-search methods, then coding efficiency is improved, but computational complexity increases significantly

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

Solution Approach 1:

The patent divides the motion estimation process into two distinct stages: integer-pel resolution search followed by sub-pel resolution refinement. This segmentation allows the computationally intensive integer-pel search to be performed first, establishing a coarse motion vector, and then a more efficient context-based refinement is applied only to determine the sub-pel position. This segmentation reduces overall computational complexity while maintaining precision.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary integer-pel resolution motion estimation before conducting sub-pel refinement. By establishing the integer-pel motion vector first, the subsequent sub-pel search can be constrained to a smaller neighborhood around this preliminary result, significantly reducing the number of candidate positions that need to be evaluated and thus lowering computational complexity.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If full search space evaluation is performed for sub-pel refinement, then optimal motion vector is guaranteed, but processing time increases

Engineering Contradiction:
Improvemotion vector optimalityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies context-based adaptive refinement where the search strategy and refinement process are tailored to local characteristics of the video content and motion patterns. By analyzing local context information and adapting the refinement process accordingly, the patent achieves near-optimal motion vectors with significantly reduced processing time compared to exhaustive full-search methods.

Inventive Principle:
Principle #3Local quality

3Productivity

If high-precision sub-pel motion estimation is implemented, then coding efficiency improves, but real-time encoding capability deteriorates

Engineering Contradiction:
Improvecoding efficiencyVSAvoidencoding speed
Core Design Contradiction:
ProductivityVSSpeed

Solution Approach 1:

The patent implements a dynamic two-stage motion estimation approach where the first stage performs integer-pel search and the second stage performs context-based sub-pel refinement. This dynamic process adapts to different video content characteristics, achieving high coding efficiency when needed while maintaining real-time encoding capability by avoiding unnecessary computational overhead in all cases.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3314894B1Refinement of a low-pel resolution motion estimation vector
Publication Date: 2024.03.20 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3314894B1 patent drawingFigure 1
  • EP3314894B1 patent drawingFigure 2
  • EP3314894B1 patent drawingFigure 3

AI summary

The refinement of a lower-pel resolution motion estimation vector is made more effective by preceding the actual refinement with a determination of a context out of a set of context candidates based on coding costs of a set of lower-pel resolution vectors within a neighborhood of the lower-pel resolution motion estimation vector, wherein each context candidate is associated with a ranking among a set of higher-pel resolution candidate vectors surrounding the lower-pel resolution motion estimation vector. The actual refinement of the lower-pel resolution motion estimation vector to obtain the higher-pel resolution motion estimation vector may thus use the ranking associated with the context determined: sequentially arranged accordingly, the set of higher-pel resolution candidate vectors may be tested sequentially.