Motion Vector Prediction via Reference Zone Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image encoding and decoding methods, such as those in the H.264/MPEG-4 AVC standard, suffer from poor video quality due to inadequate prediction of motion vectors, particularly in inter coding, as they do not effectively utilize spatial and temporal correlations between images, leading to inaccurate motion vector predictions.

Innovation Solution

A method that re-partitions reference macroblocks to better match the shape, size, and arrangement of current partitions, allowing for more precise prediction of motion vectors by using motion vectors from closer, similarly sized, and positioned reference partitions, thereby improving prediction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If standard predictive coding is used to code current macroblocks using only spatial correlations from causal reference zones, then the encoding process is simple and reliable, but the video quality is poor because temporal correlations between images are not utilized

Engineering Contradiction:
Improveencoding reliabilityVSAvoidvideo quality
Core Design Contradiction:
ReliabilityVSManufacturing precision

Solution Approach 1:

The patent segments the reference causal zone into multiple sub-zones (first causal zone with first reference macroblocks, second causal zone with second reference macroblocks) and processes different types of correlations (spatial and temporal) separately. This segmentation allows the system to combine both spatial and temporal prediction methods, improving video quality while maintaining encoding reliability through structured processing.

Inventive Principle:
Principle #1Segmentation

2Device complexity

If motion vectors are predicted using reference macroblocks of different sizes and positions, then the prediction process is simple, but the prediction precision is poor due to large spatial distances and mismatched geometries

Engineering Contradiction:
Improveprediction process complexityVSAvoidmotion vector prediction precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent applies local quality by selecting reference macroblocks with specific characteristics (same or similar size, specific positions relative to the current macroblock) for different prediction modes. For intra prediction, it uses reference macroblocks of the same size from the first causal zone. For inter prediction, it uses reference macroblocks from the second causal zone with appropriate spatial relationships. This localized selection optimizes prediction precision for each specific case while keeping the overall process manageable.

Inventive Principle:
Principle #3Local quality

3Measurement precision

If the reference causal zone is divided into multiple zones with different reference macroblocks, then prediction precision improves, but the device complexity and processing steps increase

Engineering Contradiction:
Improvemotion vector prediction precisionVSAvoidencoding device complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The reference causal zone is segmented into a first causal zone containing first reference macroblocks and a second causal zone containing second reference macroblocks. This segmentation enables the system to handle different prediction scenarios (intra and inter prediction) with dedicated reference zones, improving precision while organizing complexity into manageable structural components.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic selection of prediction modes (intra prediction mode and inter prediction mode) based on the specific coding requirements. The system can switch between using first reference macroblocks for intra prediction and second reference macroblocks for inter prediction, allowing flexible adaptation to different scenarios while managing complexity through conditional processing.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP2377323B1Image prediction by subdivision of causal regions of reference and coding using such prediction
Publication Date: 2019.09.25 ORANGE SA
  • EP2377323B1 patent drawingFigure 1A~1B
  • EP2377323B1 patent drawingFigure 2
  • EP2377323B1 patent drawingFigure 3

AI summary

A method is provided for decoding a data stream representative of an image or of a sequence of images, the stream comprising data representative of at least one portion of one of the images, the method implementing a step of predicting a motion vector of a partition of a current image portion, with respect to a reference causal zone which has been partitioned according to a mode of coding determined for the zone. The prediction step comprises, in relation to said current partition: partitioning anew at least one portion of the reference causal zone into a plurality of reference partitions, determining motion vectors respectively associated with reference partitions obtained subsequent to the new partitioning, determining a predictor motion vector on the basis of said determined motion vectors, and predicting the motion vector of said current partition on the basis of said determined predictor motion vector.