CABAC Context Model Selection for Hybrid Video Coding Tools
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards face inefficiencies in encoding and decoding due to the high algorithmic complexity and varying applicability of coding tools across different block sizes and aspect ratios, leading to suboptimal compression performance and increased bitrate.
Innovation Solution
Adopting a use-case driven context model selection for CABAC (Context Adaptive Binary Arithmetic Coding) by introducing additional context models based on block aspect ratios to selectively encode/decode tool flags, allowing encoders to skip unnecessary tool evaluations and reduce signaling overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single context model is used for encoding tool flags across all block types, then the CABAC engine complexity is reduced, but compression efficiency deteriorates due to inability to adapt to varying statistics of different block aspect ratios
Solution Approach 1:
The patent segments the context modeling process by creating multiple context models (first context model and second context model) differentiated by block aspect ratio characteristics. This segmentation allows each context model to specialize in encoding specific block types, improving compression efficiency without significantly increasing overall system complexity.
Solution Approach 2:
The patent applies local quality by selecting different context models based on local block characteristics (aspect ratio). The first context model is used for blocks with aspect ratio > 2, while the second context model is used for other blocks, allowing the encoding process to adapt to local statistical properties and improve compression efficiency.
2Measurement precision
If coding tools are evaluated for all block types regardless of applicability, then coding tool selection accuracy is improved, but algorithmic complexity increases
Solution Approach 1:
The patent changes the parameter used for context model selection from generic block properties to specific aspect ratio thresholds. By using aspect ratio as a discriminating parameter, the system can accurately determine tool applicability and select appropriate context models, improving coding tool selection accuracy while avoiding unnecessary evaluations.
Solution Approach 2:
The patent performs preliminary classification of blocks based on aspect ratio before proceeding with tool flag encoding. This preliminary action allows the system to pre-select the appropriate context model, avoiding unnecessary tool evaluations for blocks where certain tools are not applicable, thus reducing algorithmic complexity.
3Adaptability or versatility
If context models are updated frequently to adapt to current probability, then adaptability is improved, but processing speed decreases due to increased update overhead
Solution Approach 1:
The patent segments the context model update process by maintaining separate update mechanisms for different context models. Each context model can be updated independently based on its specific usage patterns, allowing frequent updates for actively used models while reducing update frequency for less critical models, thus balancing adaptability and processing speed.
Data Source
AI summary
An apparatus includes an encoder, which receives image or video data, encodes the data and provides a bitstream. The encoder includes a CABAC encoder, which receives a binary valued syntax element associated with a block of the data to be encoded, and encodes, using a selected context model, the syntax element into coded bits for the bitstream. The syntax element includes a tool flag indicating whether or not certain coding tool is employed when encoding a block of the data. For a block of the data which exhibits an aspect ratio larger than 2 and for a block of the data which exhibits an aspect ratio less than or equal to 2, and to which blocks the certain coding tool is applicable, first and second context models for encoding the tool flag are selected from a group of one or more first and second context models, respectively.