Decoder Motion Estimation via Template Matching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding techniques, such as H.264, face inefficiencies in motion prediction due to limited comparison of pixel intensity between partitions, leading to inaccurate predictions and increased residual texture data transmission costs.

Innovation Solution

A method for decoding that refines motion estimation by determining a causal partition similar in texture to the reconstructed partition, using motion vectors to adjust and refine the prediction, thereby optimizing precision and reducing residual data transmission.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If template matching is implemented at the decoder to reproduce motion prediction without transmitting motion information, then signaling cost is reduced, but prediction accuracy deteriorates due to limited pixel intensity comparison

Engineering Contradiction:
Improvesignaling costVSAvoidprediction accuracy
Core Design Contradiction:
Loss of informationVSMeasurement precision

Solution Approach 1:

The patent copies the encoder's motion prediction process to the decoder by implementing template matching that reproduces the same causal partition selection. The decoder performs identical pixel intensity comparisons and texture similarity assessments that the encoder performs, ensuring both sides arrive at the same motion prediction without requiring transmission of motion vectors or template data.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent performs preliminary action by pre-defining multiple causal partitions at the decoder before actual decoding occurs. These causal partitions are positioned in predetermined directions relative to the current partition, and their pixel intensities are pre-computed and stored in buffers, ready for rapid comparison during template matching without requiring real-time computation.

Inventive Principle:
Principle #10Preliminary action

2Device complexity

If causal partitions are selected only from predetermined directions, then search complexity is reduced, but prediction accuracy deteriorates due to inaccurate causal partition selection

Engineering Contradiction:
Improvesearch complexityVSAvoidprediction accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the search space into multiple directional zones, each with its own causal partition candidates. Instead of searching the entire frame, the search is divided into specific angular sectors around the current partition, with predetermined causal partitions assigned to each sector. This segmentation maintains low complexity while improving accuracy by ensuring relevant causal partitions in different directions are considered.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes parameters by allowing the set of predetermined directions to be configurable and adaptive. Different video content types can utilize different directional configurations, and the system can switch between fixed directional search and expanded search patterns based on motion characteristics detected in the current frame, optimizing the balance between complexity and accuracy.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2761871B1Decoder side motion estimation based on template matching
Publication Date: 2018.04.25 ORANGE SA
  • EP2761871B1 patent drawingFigure 1
  • EP2761871B1 patent drawingFigure 2
  • EP2761871B1 patent drawingFigure 3

AI summary

The invention relates to the decoding of a data signal representative of at least one current image (IN) split into partitions, said method comprising, for at least one current partition (B) to be reconstructed in said image, the steps of predicting (D3) said current partition on the basis of at least one already decoded causal partition, then reconstructing (D4) said current partition on the basis of said predicted current partition, delivering a reconstructed current partition (B). There is moreover carried out the determination (D5) of at least one already decoded causal partition (B*) with the aid of a motion estimation applied to said reconstructed current partition, then the modification (D6) of said reconstructed current partition by calculation of a function applied both to said determined causal partition (B*) and to said reconstructed current partition, delivering a modified reconstructed current partition (B). A corresponding coding of at least one image (IN) split into partitions is also implemented.