Arithmetic Coding Model Switching for Faster Probability Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entropy coding methods, such as VLC and arithmetic coding, face challenges in accurately and efficiently representing symbol probabilities in real-time due to initial probability estimation being either constant or delayed, leading to a convergence time before matching real probabilities.
Innovation Solution
An arithmetic encoding method that dynamically updates the probability model by switching between multiple probability models, including a default model, based on coding costs and switching points, allowing for rapid adaptation to real probabilities and efficient coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single probability model is used and updated during coding, then the model adapts to the source signal characteristics, but there is a convergence time delay before the estimated probability matches the real probability
Solution Approach 1:
The patent implements dynamic probability model switching by maintaining multiple probability models and selecting among them based on current signal characteristics. Instead of a single static or gradually updating model, the system dynamically transitions between models to immediately adapt to changing source signal statistics, eliminating the convergence delay inherent in single-model approaches.
Solution Approach 2:
The patent segments the probability modeling function by dividing it into multiple specialized probability models, each potentially optimized for different signal characteristics or regions. This segmentation allows the system to select the most appropriate model for the current coding context, achieving rapid adaptation without requiring a single model to converge slowly across all possible signal variations.
2Ease of manufacture
If VLC tables are defined by off-line analysis, then the coding tables are optimized for the signal characteristics, but the estimated probability is constant and may be distant from the real probability during coding
Solution Approach 1:
The patent replaces the static VLC table approach with dynamic probability model selection. While VLC tables are prepared offline, the patent enhances this by maintaining multiple probability models that can be dynamically selected during coding to match current signal characteristics, thereby maintaining the preparation simplicity of VLC while achieving the adaptability needed for accurate probability estimation.
Solution Approach 2:
The patent changes the parameter of probability estimation from a constant value (in VLC) to a selectable set of values (multiple probability models). By allowing the system to choose among different probability models based on current signal conditions, the patent maintains the offline preparation benefit while significantly improving the accuracy of probability estimation during actual coding operations.
3Productivity
If multiple probability models are maintained and switching is performed, then the convergence to real probability is faster, but the device complexity increases
Solution Approach 1:
The patent segments the probability modeling task into multiple specialized models, each handling specific signal characteristics. This segmentation improves coding efficiency by matching the right model to the right signal condition, while the modular nature of segmented models makes the overall system more manageable despite the increased number of models.
Solution Approach 2:
The patent implements self-service mechanisms where the system automatically selects and switches between probability models based on inherent signal characteristics without requiring complex external control. The probability models themselves contain the information needed to determine their applicability, allowing the system to self-manage the complexity of multiple models through automated, context-driven selection.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The invention relates to a method for arithmetic encoding of symbols in a stream (F), comprising encoding (106) a current symbol (Sc) with a current probability model (Pc), and updating (108) the current probability model according to the encoding (106) of the current symbol (Sc). The method further comprises, in the switching points (100) distributed in said stream (F), selecting (102) a probability model from a set (L) of at least two probability models; encoding (104) an identifier of the selected probability model; and adding (105) in a subset (E) the current probability model, the selected probability model (Pm) becoming the current probability model (Pc) for encoding symbols in blocks following the switching point. The invention also relates to a corresponding method for arithmetic decoding.