Intra Prediction Mode Split Using Decoder-Derived Gradient Histograms

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing backward intra prediction modes lack a solution for determining the best mode split of a block with varying directionality, leading to inefficiencies in video coding.

Innovation Solution

A method for determining a final seed intra prediction mode and mode split using a histogram of oriented gradients, allowing for improved intra prediction by partitioning blocks based on cumulative gradients and minimizing distortion or rate-distortion metrics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If backward estimation methods are used to determine multiple prediction modes for a block, then prediction accuracy is improved, but the ability to determine optimal mode split is lost

Engineering Contradiction:
Improveprediction accuracyVSAvoidmode split determination capability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent applies segmentation by dividing the current block into multiple sub-blocks and determining prediction modes for each sub-block independently using backward estimation. This allows the system to capture local directional variations within the block while maintaining the benefits of backward estimation methods.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by allowing different prediction modes to be applied to different sub-blocks based on their local characteristics. The histogram of oriented gradients is computed separately for each sub-block, enabling the system to adapt to local directional variations and optimize prediction accuracy for each region.

Inventive Principle:
Principle #3Local quality

2Productivity

If multiple prediction modes are used for a block, then compression efficiency is improved, but signaling overhead increases

Engineering Contradiction:
Improvecompression efficiencyVSAvoidsignaling overhead
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent applies self-service by deriving prediction modes automatically from the histogram of oriented gradients computed from already-decoded causal samples. This eliminates the need for explicit signaling of multiple prediction modes, as the decoder can independently determine the same modes that the encoder uses, thereby reducing signaling overhead while maintaining compression efficiency.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If blocks are split into multiple partitions with different prediction modes, then prediction accuracy is improved, but computational complexity increases

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

Solution Approach 1:

The patent applies preliminary action by computing the histogram of oriented gradients from already-decoded causal samples before determining the prediction modes for the current block. This preliminary computation of gradient histograms provides the necessary information for mode determination without requiring additional complex computations during the prediction phase, thereby reducing overall computational complexity.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4672743A1Intra prediction mode split with decoder side intra mode derivation
Publication Date: 2025.12.31 INTERDIGITAL CE PATENT HOLDINGS SAS
  • EP4672743A1 patent drawingFigure 1
  • EP4672743A1 patent drawingFigure 2
  • EP4672743A1 patent drawingFigure 3

AI summary

A method for decoding comprising: obtaining (1101) a current block; obtaining (1102) a neighbor template of samples in a reconstructed neighborhood of the current block; determining (1104) at least one seed intra prediction mode from an histogram of oriented gradients wherein each bin represent a cumulative gradient for a given angular intra prediction mode computed on the neighbor template; determining (1106) a final seed intra prediction mode and a final mode split for the current block based on the at least one seed intra prediction mode; and, generating a predictor of the current block by applying (1112) an intra prediction to each partition of the current block specified by the final mode split, the intra prediction being based on the final seed intra prediction mode.