Bitstream Optimization Metadata for AI-Oriented Image Decoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image compression technologies optimized for high-resolution and high-quality human vision are not suitable for machine-oriented artificial intelligence services, necessitating improved encoding/decoding methods and apparatuses for enhanced efficiency, task optimization, latency characteristics, and frequency characteristics in bitstreams.

Innovation Solution

A decoding method determines task type, latency, and frequency characteristics from a bitstream, while an encoding method encodes optimization information into the bitstream, with a recording medium storing and transmitting the encoded bitstream for decoding and task restoration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If existing image compression technologies are used, then high-resolution and high-quality image processing for human vision is achieved, but they are not suitable for artificial intelligence services

Engineering Contradiction:
Improvesuitability for artificial intelligence servicesVSAvoidimage quality
Core Design Contradiction:
Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent applies local quality by encoding different regions of the image with different compression characteristics. Important regions for AI tasks (such as regions containing objects of interest) are preserved with higher quality, while less important regions are compressed more aggressively. This is achieved through region-of-interest (ROI) based encoding where the encoder identifies and prioritizes specific areas that are critical for AI service performance.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes encoding parameters dynamically based on the intended AI task. Different compression parameters (such as quantization step sizes, transform block sizes, and prediction modes) are applied depending on the specific AI service requirements. This allows the system to optimize the balance between compression ratio and task performance by adjusting encoding parameters according to the target application.

Inventive Principle:
Principle #35Parameter changes

2Quantity of substance

If compression ratio is increased to reduce data transmission, then bandwidth efficiency is improved, but encoding/decoding efficiency deteriorates

Engineering Contradiction:
Improvebitstream sizeVSAvoidencoding/decoding efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent performs preliminary analysis of the image content and identifies regions of interest before the main compression process. This preliminary action allows the encoder to pre-determine which areas require higher preservation and which can be compressed more aggressively, avoiding the need for complex post-processing adjustments and improving overall encoding efficiency.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent employs dynamic encoding strategies where compression parameters are adjusted on a block-by-block or region-by-region basis rather than applying uniform compression across the entire image. This dynamic approach allows the system to achieve better compression ratios in less important regions while maintaining necessary quality in critical areas, thereby improving both bitstream size reduction and encoding efficiency.

Inventive Principle:
Principle #15Dynamics

3Manufacturing precision

If compression is optimized for human vision, then visual quality is improved, but task-oriented performance deteriorates

Engineering Contradiction:
Improvevisual qualityVSAvoidtask performance
Core Design Contradiction:
Manufacturing precisionVSReliability

Solution Approach 1:

The patent applies local quality by encoding different regions of the image with different compression characteristics. Important regions for AI tasks (such as regions containing objects of interest) are preserved with higher quality, while less important regions are compressed more aggressively. This is achieved through region-of-interest (ROI) based encoding where the encoder identifies and prioritizes specific areas that are critical for AI service performance.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes encoding parameters dynamically based on the intended AI task. Different compression parameters (such as quantization step sizes, transform block sizes, and prediction modes) are applied depending on the specific AI service requirements. This allows the system to optimize the balance between compression ratio and task performance by adjusting encoding parameters according to the target application.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4648413A1Encoding/decoding method and apparatus, and recording medium storing bitstream
Publication Date: 2025.11.12 LG ELECTRONICS INC
  • EP4648413A1 patent drawingFigure 1
  • EP4648413A1 patent drawingFigure 2
  • EP4648413A1 patent drawingFigure 3

AI summary

Provided are an encoding/decoding method and apparatus, and a computer-readable recording medium generated by the encoding method. The decoding method according to the present disclosure is performed by the decoding apparatus, and comprises the steps of: obtaining, from a bitstream, optimization information for performing a task; and determining at least one of a type of the task, a latency characteristic of the task, or a frequency characteristic of information encoded in the bitstream, on the basis of the optimization information.