Reference-Channel Feature Decoding for Large-Scale AI Image Data
Find Innovative SolutionsGenerate 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
Engineering 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
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
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
2Measurement precision
If reference relationships among channels are utilized for prediction, then prediction accuracy is improved, but device complexity increases
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
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
Data Source
Figure 1
Figure 2
Figure 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.