Adaptive Transform Kernel Selection for Video Intra Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding technologies face challenges in achieving optimal compression efficiency due to the complexity of managing multiple reference lines for intra prediction, transform partitioning, and transform kernels, which affects the balance between compression ratio and distortion tolerance.

Innovation Solution

A harmonized design is implemented that allows for the selection of different sets of transform kernels and partition types based on the reference lines used for intra prediction, enabling adaptive processing of video frames by distinguishing between adjacent and non-adjacent reference lines to optimize transform operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple reference lines are used for intra prediction, then prediction accuracy is improved, but processing complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the transform processing by creating different transform kernel sets (first set and second set) corresponding to different reference line types. This segmentation allows the system to handle adjacent and non-adjacent reference lines differently, reducing the overall processing complexity while maintaining prediction accuracy for each case.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different transform kernels locally based on the reference line type. Adjacent reference lines use one set of transform kernels while non-adjacent reference lines use another set. This local differentiation optimizes processing efficiency for each specific case without compromising the overall prediction accuracy.

Inventive Principle:
Principle #3Local quality

2Productivity

If transform partitioning is applied, then compression efficiency is improved, but computational complexity increases

Engineering Contradiction:
Improvecompression efficiencyVSAvoidcomputational complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces dynamic adaptability by allowing the transform kernel set to change based on the reference line type. The system dynamically selects between the first transform kernel set and the second transform kernel set, enabling flexible optimization of compression efficiency while managing computational complexity through conditional processing.

Inventive Principle:
Principle #15Dynamics

3Reliability

If adaptive transform operations are used, then distortion tolerance is improved, but processing overhead increases

Engineering Contradiction:
Improvedistortion toleranceVSAvoidprocessing overhead
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent changes the transform kernel parameters based on the reference line type. By adjusting which transform kernel set is applied (first set for adjacent, second set for non-adjacent), the system adapts to different prediction scenarios, improving distortion tolerance while keeping processing overhead manageable through parameter-based differentiation rather than complete reprocessing.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS12108031B2Harmonized design among multiple reference line intra prediction, transform partitioning and transform kernels
Publication Date: 2024.10.01 TENCENT AMERICA LLC
  • US12108031B2 patent drawing
  • US12108031B2 patent drawing
  • US12108031B2 patent drawing

AI summary

This disclosure relates to a harmonized design among multiple reference line intra prediction, transform partitioning, and transform kernels or processing video information. One example method is disclosed. The method includes retrieving a data block of a video frame, the data block having an adjacent reference line and one or more non-adjacent reference lines in the video frame; determining whether one of the one or more non-adjacent reference lines is to be used for intra prediction of the data block. When an adjacent reference line is used for the intra prediction of the data block, selecting a target transform kernel from a first set of transform kernels to transform the data block; and when a non-adjacent reference line is used for the intra prediction of the data block, selecting the target transform kernel from a second set of transform kernels to transform the data block.