Reference-Channel Feature Decoding for Large-Scale AI Image Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image compression technologies are not optimized for machine-oriented tasks and lack efficiency in processing large volumes of image data required for artificial intelligence services.

Innovation Solution

A feature encoding/decoding method that predicts channels based on reference relationships among channels, utilizing a bitstream for encoding and decoding, and storing the bitstream in a recording medium to enhance encoding/decoding efficiency and prediction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If existing image compression technologies are used for machine-oriented tasks, then high-resolution and high-quality image processing is achieved, but encoding/decoding efficiency is insufficient for processing large volumes of image data

Engineering Contradiction:
Improveencoding/decoding efficiencyVSAvoidvolume of image data
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent segments the image data processing by separating different color channels (e.g., luminance and chrominance channels) and applying different compression strategies to each channel based on their specific characteristics and machine task requirements, thereby improving overall encoding/decoding efficiency for large volumes of image data

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different quality levels and compression parameters to different regions or channels of the image data according to their importance for machine tasks, allowing more critical channels to maintain higher quality while less critical channels use more aggressive compression, thus improving efficiency without sacrificing essential information

Inventive Principle:
Principle #3Local quality

2Measurement precision

If reference relationships among channels are utilized for prediction, then prediction accuracy is improved, but device complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidencoding/decoding complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary analysis and establishment of reference relationships among channels during the encoding phase, pre-computing prediction models and storing them for reuse during decoding, thereby reducing the computational complexity at decoding time while maintaining high prediction accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses reference channels as templates or copies to predict target channels, leveraging the similarity and correlation between channels to generate accurate predictions without requiring complex processing, thus improving prediction accuracy while controlling device complexity

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4651496A1Channel prediction-based feature encoding/decoding method and device, and recording medium storing bitstream
Publication Date: 2025.11.19 LG ELECTRONICS INC
  • EP4651496A1 patent drawingFigure 1
  • EP4651496A1 patent drawingFigure 2
  • EP4651496A1 patent drawingFigure 3

AI summary

A feature encoding/decoding method and device, and a computer-readable recording medium produced by the feature encoding method are provided. The feature decoding method according to the present disclosure is a feature decoding method performed by the feature decoding device and may comprise the steps of: obtaining reference information from a bitstream; determining at least one reference channel for a target channel on the basis of the reference information; and predicting the target channel on the basis of the determined reference channel.