Multi-Model Reference Filtering for Illumination-Compensated Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently addressing illumination variations between reference and current blocks, leading to significant prediction errors, particularly in inter prediction methods.
Innovation Solution
Implementing a multiple-tap filter model for illumination compensation (MPRF) to classify reference samples into groups based on thresholds, applying different filters to these groups, and using these filters to generate prediction blocks, thereby reducing illumination-related errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional inter prediction methods are used, then coding efficiency is maintained at acceptable levels, but prediction errors increase significantly due to illumination variations
Solution Approach 1:
The patent segments the reference block into multiple regions based on illumination characteristics. Reference samples are classified into different groups (e.g., first group and second group) based on their illumination levels, and different filter models are applied to each group. This segmentation allows the system to handle illumination variations locally rather than uniformly, improving prediction accuracy without significantly impacting overall coding efficiency.
Solution Approach 2:
The patent applies different filter models to different regions of the reference block based on their specific illumination characteristics. Instead of using a single filter model for the entire block, the system selects appropriate filter models (e.g., first filter model for first group, second filter model for second group) that are optimized for each local region's illumination conditions. This local quality approach enables precise compensation for illumination variations while maintaining coding efficiency.
2Reliability
If multiple filter models are applied to different reference sample groups, then illumination compensation improves, but processing complexity increases
Solution Approach 1:
The patent performs preliminary classification of reference samples into different groups based on illumination characteristics before applying the filter models. This preliminary action organizes the reference samples in advance, allowing the system to select and apply appropriate filter models systematically. By preparing the reference sample groups beforehand, the system reduces the complexity of the subsequent filtering process while maintaining accurate illumination compensation.
Solution Approach 2:
The patent changes parameters such as filter model selection and classification thresholds to optimize the balance between illumination compensation accuracy and processing complexity. The system adjusts these parameters based on the specific characteristics of the reference block and can configure the number of groups and filter models to match the actual illumination variation needs, preventing unnecessary complexity when simple illumination conditions are present.
3Measurement precision
If reference samples are classified into multiple groups with different filters, then prediction precision improves, but computational requirements increase
Solution Approach 1:
The patent applies partial action by classifying reference samples into groups and applying filters only where necessary. Instead of uniformly applying complex filtering to all reference samples, the system identifies regions with significant illumination variations and applies appropriate filter models only to those regions. This partial action approach maintains high classification precision for problematic areas while reducing computational energy consumption in regions with uniform illumination.
Solution Approach 2:
The patent implements dynamic adaptation by adjusting the number of reference sample groups and filter model applications based on the actual illumination characteristics of the current block. The system can dynamically configure the classification thresholds and filter selection based on the degree of illumination variation detected, allowing it to use higher precision classification when needed and simpler processing when illumination is uniform, thus optimizing the balance between precision and computational energy.
Data Source
AI summary
A video coder (encoder or decoder) classifies, based on a first threshold, first samples neighboring a reference block into a first and a second group. The coder classifies, based on a second threshold, second samples corresponding to the first samples and neighboring a current block into the first and second groups. A first filter is determined based on first sample pairs of samples, of the first samples, classified in the first group and corresponding samples, of the second samples, classified in the first group. A second filter is determined based on second sample pairs of samples, of the first samples, classified in the second group and corresponding samples, of the second samples, classified in the second group. The current block is coded based on a prediction block determined by applying at least the first and second filters to the reference block.