Adaptive Feature Encoding And Decoding For AI Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies are unsuitable for artificial intelligence services due to their focus on high-resolution, high-quality image processing for human vision, lacking efficiency and adaptability for machine-oriented tasks.
Innovation Solution
A feature encoding/decoding method and apparatus that determines the type and property of the input source, incorporating supplementary and meaning information, and provides privacy protection, enabling efficient encoding/decoding of feature data for machine tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image compression technology is used, then high-resolution and high-quality image processing for human vision is achieved, but it is unsuitable for artificial intelligence services
Solution Approach 1:
The patent introduces dynamic adaptation by detecting whether the input is natural image data or feature data and switching encoding modes accordingly. The encoding apparatus dynamically adjusts its operation based on input type identification, enabling it to adapt between human-vision-oriented and machine-task-oriented processing modes.
Solution Approach 2:
The patent changes the fundamental parameters of the encoding process by introducing feature map type information (first feature map type, second feature map type) and adjusting encoding strategies based on these parameters. This allows the system to optimize for different AI service requirements while maintaining compatibility with traditional image processing.
2Measurement precision
If image compression technology optimized for human vision is used, then high-quality images are produced, but encoding/decoding efficiency for machine tasks is reduced
Solution Approach 1:
The patent segments the encoding process into distinct pathways based on input type. By dividing the encoding logic into natural image encoding and feature data encoding branches, the system can apply optimized strategies for each type, improving overall efficiency for machine tasks while preserving quality for human vision applications.
Solution Approach 2:
The patent introduces an intermediary mechanism (input type detection and feature map type identification) that mediates between the encoding apparatus and the input data. This intermediary layer enables the system to select appropriate encoding strategies, thereby improving productivity for AI services without sacrificing image quality when needed.
3Productivity
If feature data encoding is implemented, then encoding efficiency for AI services is improved, but complexity of determining input type and feature map type increases
Solution Approach 1:
The patent applies preliminary action by performing input type detection and feature map type identification before the actual encoding process. By determining the input type in advance and preparing the appropriate encoding strategy beforehand, the system reduces complexity during the main encoding operation and improves overall efficiency.
Solution Approach 2:
The encoding apparatus performs self-service by automatically detecting input types and selecting appropriate encoding modes without requiring external intervention. The system self-determines whether to process natural images or feature data and adjusts its operation accordingly, managing the complexity internally while maintaining simplicity for users.
4Measurement precision
If feature map type information is encoded, then decoding accuracy is improved, but amount of encoding information increases
Solution Approach 1:
The patent applies partial action by selectively encoding feature map type information only when necessary for accurate decoding. Rather than encoding all possible information, the system encodes only the essential feature map type identifiers (first or second feature map type) that are critical for maintaining decoding accuracy, thereby balancing information volume with precision.
Data Source
AI summary
Provided are a feature encoding/decoding method and device, a recording medium on which a bitstream generated by the feature encoding method is stored, and a method for transmitting the bitstream. The feature decoding method according to the present disclosure comprises the steps of: determining whether or not an input source is feature data; determining a feature map type of the feature data based on the input source being the feature data; obtaining encoding information of the feature data from a bitstream; and decoding the feature data based on the feature map type and the encoding information, and whether or not the input source is the feature data may be determined based on input format identification information that is obtained from a sequence parameter set.