Cross-Color LM Intra Prediction at Restricted Block Boundaries

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Intra prediction methods for chroma components in video coding face challenges with inaccurate LM parameters due to unavailable neighboring samples at boundaries, leading to inefficiencies and increased buffer requirements.

Innovation Solution

The method employs additional neighboring samples from accessible areas to derive LM parameters when traditional samples are inaccessible, and adapts multi-LM modes to handle restricted boundaries, ensuring accurate prediction by using expanded or modified sample regions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional LM parameter derivation uses only immediately adjacent neighboring samples, then buffer requirements are reduced, but prediction accuracy deteriorates due to unavailable samples at boundaries

Engineering Contradiction:
Improveprediction accuracyVSAvoidbuffer requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extends the sample selection from a single dimension (immediately adjacent samples) to multiple dimensions by incorporating samples from both top and left neighboring blocks, effectively using a two-dimensional region around the current block to derive LM parameters, thereby improving accuracy without excessive buffer increase

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent performs preliminary derivation of LM parameters using available neighboring samples before actual prediction, allowing the system to adapt to boundary conditions in advance and select appropriate parameter sets based on sample availability, thus maintaining accuracy while managing buffer requirements

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If additional neighboring samples are used to derive LM parameters, then prediction accuracy improves, but device complexity increases

Engineering Contradiction:
Improveparameter derivation accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements dynamic selection of LM parameter sets based on the availability of neighboring samples at block boundaries. The system adapts its behavior according to position (top boundary, left boundary, or interior), automatically selecting the appropriate parameter derivation method without requiring complex manual configuration

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies different LM parameter derivation strategies to different regions of the picture based on local boundary conditions. Each block independently selects its parameter set based on its specific position and available neighboring samples, allowing localized optimization without increasing global system complexity

Inventive Principle:
Principle #3Local quality

3Reliability

If restricted boundaries are handled by using only available samples, then buffer requirements are reduced, but prediction reliability deteriorates

Engineering Contradiction:
Improveprediction reliabilityVSAvoidsample availability
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent changes the parameters (LM parameters a and b) based on the available neighboring samples. When top samples are unavailable, it uses left samples with adjusted parameter sets, and vice versa. This parameter adaptation ensures reliable prediction across different boundary conditions by dynamically adjusting to sample availability

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent creates multiple copies of LM parameter sets (first LM parameter set, second LM parameter set) derived from different neighboring sample configurations. These parameter copies are stored and selected based on boundary conditions, ensuring reliable prediction without requiring all original samples to be simultaneously available

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3785437B1Method and apparatus for restricted linear model parameter derivation in video coding
Publication Date: 2026.04.01 HFI INNOVATION INC
  • EP3785437B1 patent drawingFigure 1~2B
  • EP3785437B1 patent drawingFigure 3A~4A
  • EP3785437B1 patent drawingFigure 4B~5B

AI summary

Methods and apparatus for cross-color LM (Linear Mode) Intra prediction are disclosed. According to one method, when a current top area or a current left area of the current block is across a restricted boundary, a current LM parameter set is derived based on reduced neighboring reconstructed areas. The reduced neighboring reconstructed areas comprise a reduced top area having less than N top lines or less than N left lines if the current top area or the current left area is across the restricted boundary. According to another method, if the current top area is across the restricted boundary, expanded left area is used for deriving LM parameter set and if the current left area is across the restricted boundary, expanded top area is used for deriving LM parameter set. A method for multi-LM mode is also disclosed for restricted boundaries.