Parallel Entropy Encoding with Shared Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video encoding methods, such as H.264/AVC, suffer from reduced compression efficiency due to independent processing of slices, which leads to unbalanced computing loads and inefficient entropy encoding across different image regions.
Innovation Solution
The method involves subdividing images into regions and using parallel encoding branches that share a common set of probability models updated at regular intervals, allowing for simultaneous entropy encoding across multiple branches, thereby accounting for shared statistical dependencies and improving encoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If images are divided into slices for independent processing, then encoding speed is improved through parallel processing, but compression efficiency deteriorates due to loss of statistical dependencies between slices
Solution Approach 1:
The image is divided into multiple slices that can be processed in parallel, with each slice further divided into entropy slices. This segmentation enables parallel processing while maintaining manageable data units for entropy encoding operations.
Solution Approach 2:
Multiple probability models from different slices are merged into a single combined probability model. The combined model aggregates frequency information from all slices to create a unified statistical representation that captures dependencies across the entire image, resolving the loss of inter-slice statistical information.
2Ease of operation
If separate probability models are used for each slice, then encoding independence is improved for parallel processing, but encoding efficiency deteriorates due to unbalanced computing loads and loss of shared statistics
Solution Approach 1:
The encoding process is segmented into independent entropy slices within each slice, allowing parallel processing while maintaining the ability to combine results. Each entropy slice can be processed independently but contributes to the overall combined probability model.
Solution Approach 2:
A single combined probability model serves multiple slices simultaneously, making the model universal across different image regions. This multi-functional model improves encoding efficiency by providing consistent statistical information across all slices while still allowing independent processing.
3Measurement precision
If adaptive probability model updating is performed continuously, then encoding precision is improved through context adaptation, but computing load increases and processing speed decreases
Solution Approach 1:
Probability model updates are performed periodically at predetermined intervals rather than continuously after every symbol. This periodic updating maintains encoding precision through context adaptation while significantly reducing the computational overhead and increasing processing speed.
Solution Approach 2:
Instead of updating probability models for every single symbol (excessive action), the system performs partial updates at selected intervals. This partial updating approach provides sufficient context adaptation for good encoding precision while avoiding the excessive computational burden of continuous updates.
Data Source
AI summary
Images are divided into image regions. The image regions are processed in encoding cycles such that during an encoding cycle, entropy encoding is carried out in a plurality of parallel encoding branches. In each encoding branch the symbols of an image region are encoded by entropy encoding, which is based on one or more probability models, which take into account the frequencies of symbols occurring in image regions. The probability models are adapted on the basis of the symbols occurring in the image region. A common set of probability models is updated at predetermined time intervals on the basis of frequencies that are adapted in at least one temporally preceding encoding cycle.