Image Processing Using Original Pixels for Matrix Intra Prediction

Solution Overview

Problem

The existing methods for generating predicted images in video coding, such as those described in JVET-N0217, rely on averaging pixels and using these averaged values for interpolation, which can lead to decreased prediction accuracy due to reduced correlation between the prediction block and the predicted image.

Innovation Solution

The proposed method involves using the original pixel of the reference image as the top adjacent pixel for interpolation during intra prediction, rather than averaged pixels, to improve prediction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If averaged pixels are used for interpolation processing, then the processing is simplified, but the prediction accuracy deteriorates

Engineering Contradiction:

Improveinterpolation processing simplicityVSAvoidprediction accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent applies local quality by using different pixel sources for different positions in the predicted image. Specifically, original pixels are used for the top adjacent pixel position while averaged pixels are used for other positions, allowing each region to have the quality appropriate to its functional requirements for maintaining prediction accuracy.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments the interpolation processing into different parts: one part uses original pixels from the reference image for the top adjacent pixel, while another part uses averaged pixels for the remaining pixels. This segmentation allows the system to optimize each part differently to achieve both simplicity and accuracy.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If original pixels are used for interpolation processing, then the prediction accuracy is improved, but the processing complexity increases

Engineering Contradiction:

Improveprediction accuracyVSAvoidinterpolation processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by using different pixel sources for different positions in the predicted image. Specifically, original pixels are used for the top adjacent pixel position while averaged pixels are used for other positions, allowing each region to have the quality appropriate to its functional requirements for maintaining prediction accuracy.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments the interpolation processing into different parts: one part uses original pixels from the reference image for the top adjacent pixel, while another part uses averaged pixels for the remaining pixels. This segmentation allows the system to optimize each part differently to achieve both simplicity and accuracy.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12445644B2Image processing device and image processing method

Publication Date: 2025.10.14 SONY GROUP CORP

AI summary

The present technology relates to an image processing device and an image processing method capable of improving prediction accuracy.A predicted image of a current prediction block is generated by performing interpolation processing using an original pixel of a reference image as a top adjacent pixel adjacent to a top of the predicted image of the current prediction block to be encoded/decoded when performing an intra prediction using a matrix operation. Then, the current prediction block is encoded/decoded using the predicted image. The present technology can be applied, for example, to a case of performing the encoding and decoding an image.