Motion Compensation Fractional Sample Interpolation Bit Depth Truncation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing motion compensation techniques for digital video encoding, such as those in H.264/AVC, face challenges in achieving high precision and low complexity due to rounding errors and high computational complexity, especially when dealing with fractional sample interpolation and bit depth truncation.

Innovation Solution

A novel motion compensation method applies a first and second fractional sample interpolation using 8-tap FIR filters with specific coefficients, truncating results to maintain constant precision, and scaling samples to a consistent bit depth for accurate quarter-pel sample generation, reducing rounding errors and computational complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple averaging operations are cascaded to generate fractional samples, then the resolution and accuracy of motion compensation is improved, but rounding errors are introduced which adversely affect accuracy and efficiency

Engineering Contradiction:
Improvemotion compensation accuracyVSAvoidrounding error accumulation
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent applies preliminary action by performing bit depth conversion of reference samples before the fractional sample interpolation process. By converting reference samples to a higher bit depth (second bit depth) prior to interpolation, the system prevents rounding errors from being introduced during subsequent averaging operations, thereby maintaining accuracy throughout the entire motion compensation process.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If fractional sample interpolation is applied to increase reference picture resolution, then motion vector accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improvemotion vector accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by establishing a fixed relationship between bit depths: the second bit depth for fractional samples is defined as the number of bits available to represent fractional samples. This parameter specification allows the system to achieve high precision motion compensation through controlled bit depth management rather than complex multi-step interpolation processes, thereby reducing computational complexity while maintaining accuracy.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If bit depth conversion is applied to reference samples, then precision of fractional sample representation is improved, but additional processing steps are required

Engineering Contradiction:
Improvefractional sample representation precisionVSAvoidprocessing steps
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies universality by designing the bit depth conversion process to serve multiple functions simultaneously. The conversion from first bit depth to second bit depth not only prepares reference samples for accurate fractional interpolation but also establishes a unified precision standard for all subsequent processing steps. This multi-functional approach eliminates the need for separate precision adjustment steps, reducing overall processing complexity while maintaining high precision.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3826307B1Method for motion compensated prediction
Publication Date: 2023.10.04 NTT DOCOMO INC
  • EP3826307B1 patent drawingFigure 1
  • EP3826307B1 patent drawingFigure 2~3
  • EP3826307B1 patent drawingFigure 4

AI summary

The invention relates to a motion compensation performed under an inter- frame prediction. A fractional sample interpolation is applied on retrieved samples to generate fractional samples. The fractional sample interpolation comprises a plurality of filtering operations, and all of filtered results from the filtering operations are truncated down to a predetermined bit depth independent of the bit depth of the samples stored in the reference picture memory.