Image Encoding Device Partial Transform for Residual Quality

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

When the similarity between reference images is low, the prediction accuracy of image encoding devices decreases, leading to large prediction residuals and degradation of image quality, which affects encoding efficiency.

Innovation Solution

An image encoding device that evaluates the similarity between reference images on a pixel-by-pixel basis and applies orthogonal transform and quantization only to specific partial areas with high prediction residuals, improving energy compaction and reducing degradation of transform coefficients.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If orthogonal transform and quantization are applied to the entire prediction residual, then encoding is performed uniformly across all areas, but quantization degradation propagates to areas with high prediction accuracy, lowering overall image quality

Engineering Contradiction:
Improveimage qualityVSAvoidquantization degradation propagation
Core Design Contradiction:
Manufacturing precisionVSLoss of information

Solution Approach 1:

The prediction residual is divided into multiple regions based on prediction accuracy evaluation. High-accuracy regions and low-accuracy regions are segmented separately, allowing different processing strategies to be applied to each region, preventing quantization degradation from propagating to high-accuracy areas.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Different quantization strategies are applied to different regions: high-accuracy regions use skip processing or lower quantization strength to preserve quality, while low-accuracy regions undergo full orthogonal transform and quantization. This local differentiation optimizes overall image quality by protecting important regions.

Inventive Principle:
Principle #3Local quality

2Loss of energy

If orthogonal transform is applied to prediction residuals with large values, then energy compaction is achieved, but high frequency components degrade due to rough quantization, and degradation propagates through inverse transform

Engineering Contradiction:
Improveenergy compactionVSAvoidtransform coefficient quality
Core Design Contradiction:
Loss of energyVSManufacturing precision

Solution Approach 1:

Orthogonal transform and quantization are applied partially only to low-accuracy regions rather than the entire residual. This partial action maintains energy compaction benefits in regions where needed while avoiding excessive quantization degradation in high-accuracy regions.

Inventive Principle:
Principle #16Partial or excessive action

3Adaptability or versatility

If prediction is performed using multiple reference images, then prediction coverage is improved, but areas with low similarity between reference images produce large prediction residuals, reducing encoding efficiency

Engineering Contradiction:
Improveprediction coverageVSAvoidencoding efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The prediction accuracy evaluation is performed preliminarily before orthogonal transform and quantization. This preliminary evaluation identifies regions that will benefit from skip processing, allowing the encoding process to adaptively optimize efficiency while maintaining prediction coverage benefits.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20240297995A1Image encoding device, image decoding device and program
Publication Date: 2024.09.05 NIPPON HOSO KYOKAI
  • US20240297995A1 patent drawing
  • US20240297995A1 patent drawing
  • US20240297995A1 patent drawing

AI summary

An image encoding device (1) encodes a block-based target image. The image encoding device (1) comprises: a predictor (109) configured to generate a prediction image corresponding to the target image by performing prediction using a plurality of reference images; an evaluator (111) configured to evaluate a degree of similarity between the plurality of reference images on a pixel-by-pixel basis; a calculator (101) configured to calculate a prediction residual indicating a pixel-based difference between the target image and the prediction image; a determiner (112) configured to determine a partial area, to which an orthogonal transform and quantization are to be applied, of the prediction residual based on a result of the evaluation by the evaluator; and a transformer/quantizer (102) configured to perform an orthogonal transform and quantization only for the partial area in the prediction residual.