MIP-Based LFNST Kernel Selection for VVC Intra Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing Low-Frequency Non-Separable Transform (LFNST) technology in H.266/VVC lacks variability in transform selection for non-traditional intra prediction modes, leading to reduced encoding efficiency and applicability, especially in Matrix-based Intra Prediction (MIP) modes.

Innovation Solution

A transform method that determines a Matrix-based Intra Prediction (MIP) parameter to select a flexible LFNST transform kernel, signaled in the video bitstream, enhancing the applicability of LFNST to non-traditional intra prediction modes and improving encoding and decoding efficiencies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If LFNST transform set selection is based on intra prediction mode, then encoding efficiency is improved for traditional modes, but applicability to non-traditional intra prediction modes deteriorates due to lack of variability

Engineering Contradiction:
Improveencoding efficiencyVSAvoidapplicability to non-traditional intra prediction modes
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent changes the selection parameter for LFNST transform kernels from intra prediction mode to MIP parameter (mipModeIdx). This allows the transform kernel selection to vary with different MIP modes, providing the necessary variability for non-traditional intra prediction modes while maintaining encoding efficiency. The patent establishes a mapping between mipModeIdx values and specific LFNST transform kernels, enabling flexible adaptation.

Inventive Principle:
Principle #35Parameter changes

2Device complexity

If transform set selection lacks variability for non-traditional modes, then device complexity is reduced, but encoding efficiency deteriorates

Engineering Contradiction:
Improvetransform set selection complexityVSAvoidencoding efficiency
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent introduces a new parameter (mipModeIdx) for transform kernel selection that maintains simple indexing logic while improving encoding efficiency. The mapping from mipModeIdx to LFNST transform kernel is implemented through straightforward index tables, avoiding complex decision logic and maintaining low computational overhead despite the increased adaptability.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4087253B1Transformation method, encoder, decoder, and storage medium
Publication Date: 2026.04.22 GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD
  • EP4087253B1 patent drawingFigure 1~2A
  • EP4087253B1 patent drawingFigure 2B~3
  • EP4087253B1 patent drawingFigure 4~6A

AI summary

Disclosed are a transformation method, an encoder, a decoder, and a storage medium. The method comprises: determining a prediction mode parameter of the current block; when the prediction mode parameter indicates that the current block determines an intra-frame prediction value by means of matrix-based intra-frame prediction (MIP), determining a MIP parameter; determining the intra-frame prediction value of the current block according to the MIP parameter, and calculating a prediction difference between the current block and the intra-frame prediction value; when the current block uses a low-frequency non-separable secondary transform (LFNST), determining, according to the MIP parameter, an LFNST transform kernel used by the current block, setting an LFNST index number, and writing same into a video code stream; and performing transformation processing on the prediction difference by using the LFNST transform kernel.