Image Coding Context Sharing Across Block Sizes for Better Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional arithmetic coding methods for video data compression do not provide sufficient coding efficiency due to the large number of contexts required, leading to inaccurate updates and decreased prediction accuracy of symbol occurrence probabilities.
Innovation Solution
An image coding method that selects a shared context for signals with similar statistical properties across different processing units, reducing the number of contexts needed and increasing the frequency of updates, thereby enhancing prediction accuracy and coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional arithmetic coding uses separate contexts for each processing unit size, then it can adapt to different image features, but the number of contexts becomes large leading to insufficient update frequency and decreased prediction accuracy
Solution Approach 1:
The patent merges contexts by having larger processing units share contexts with smaller processing units. Specifically, a processing unit of size 8×8 shares contexts with processing units of sizes 4×4, 2×2, and 1×1, reducing the total number of contexts from 48 to 24. This merging increases the update frequency for each context while maintaining adaptability to different image features through selective sharing.
Solution Approach 2:
The patent implements universality by designing contexts to serve multiple processing unit sizes simultaneously. Each context is made multi-functional so that it can be used by processing units of different sizes (1×1, 2×2, 4×4, and 8×8) depending on the statistical properties of the coefficients being coded, rather than having dedicated contexts for each size.
2Measurement precision
If the number of contexts is reduced by using shared contexts, then update frequency increases and prediction accuracy improves, but the ability to adapt to specific processing unit features may decrease
Solution Approach 1:
The patent applies local quality by allowing different levels of context sharing based on the specific needs of different processing unit sizes and coefficient positions. Not all contexts are shared equally - the sharing is selective and localized to specific situations where it provides benefit, maintaining adaptability while improving update frequency.
Solution Approach 2:
The patent implements dynamics by making context selection adaptive and flexible. The system dynamically determines which contexts to share and which to keep separate based on the statistical properties of the coefficients and the processing unit size, allowing the context structure to adapt to different image content and processing scenarios.
3Productivity
If separate contexts are used for different processing unit sizes, then coding can be optimized for each size, but memory size for storing contexts increases
Solution Approach 1:
The patent merges context storage by having multiple processing unit sizes share the same context tables. Instead of maintaining separate context storage for each processing unit size (which would quadruple the memory requirements), the system uses shared context storage that is accessed by different processing unit sizes based on their statistical properties.
4Measurement precision
If more contexts are used to capture statistical properties of different processing units, then coding accuracy may improve, but the complexity of the coding system increases
Solution Approach 1:
The patent reduces system complexity by making contexts universal and multi-functional. Instead of having dedicated contexts for each processing unit size (which would increase complexity), each context is designed to serve multiple processing unit sizes, simplifying context management while maintaining coding accuracy through selective sharing based on statistical properties.
Data Source
AI summary
An image coding method comprising: obtaining current signals to be coded of each of the processing units of the image; generating a binary signal by performing binarization on each of the current signals to be coded; selecting a context for each of the current signals to be coded from among a plurality of contexts; performing arithmetic coding of the binary signal by using coded probability information associated with the context selected in the selecting; and updating the coded probability information based on the binary signal, wherein, in the selecting, the context for the current signal to be coded is selected, as a shared context, for a signal which is included in one of a plurality of processing units and has a size different from a size of the processing unit including the current signal to be coded.