Video Decoder Bi-Prediction Using WP and GBI Coefficients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The combination of Weighted prediction (WP) and Generalized bi-prediction (GBI) is challenging due to the difficulty in using weight coefficients, and GBI prediction may not achieve the same effect as weighted prediction, especially in merge prediction modes, leading to increased table sizes and limited applicability in uni-prediction.
Innovation Solution
An image decoding apparatus that generates bi-prediction images using weight coefficients from coded data, selecting coefficients based on reference picture indices or GBI indices from a table, and applying specific weight coefficients for pixel values in reference images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Weighted prediction (WP) and Generalized bi-prediction (GBI) are combined, then prediction accuracy is improved, but the complexity of managing weight coefficients increases
Solution Approach 1:
The patent merges WP and GBI prediction methods into a unified framework where both techniques can be applied together. The decoder combines the weighted prediction results (using reference picture indices) with GBI prediction results (using GBI indices) to generate the final prediction image, thereby improving prediction accuracy while managing complexity through integration.
Solution Approach 2:
The patent creates a universal prediction framework that can handle both WP and GBI methods within the same decoding process. The system universally manages weight coefficients for both prediction types, allowing flexible application of either or both methods depending on the coding conditions, thus improving adaptability without proportionally increasing complexity.
2Adaptability or versatility
If GBI prediction is used to generate prediction images with fade or resolve effects, then prediction flexibility is improved, but the effect equivalent to weighted prediction is not achieved
Solution Approach 1:
The patent creates a composite prediction approach by combining GBI prediction with weighted prediction. Instead of relying solely on GBI's fade/resolve effects, the system layers weighted prediction on top of GBI prediction, merging the flexibility of GBI with the quality assurance of WP to achieve both flexibility and equivalent prediction effects.
3Measurement precision
If weight coefficients are decoded for each reference picture in merge prediction mode, then prediction accuracy is improved, but the inheritance of available weight coefficients from merge candidates is lost
Solution Approach 1:
The patent applies preliminary action by pre-decoding weight coefficients for each reference picture during the merge candidate preparation phase. These pre-decoded coefficients are then stored and readily available for inheritance when merge prediction mode is selected, eliminating the need for re-decoding and improving decoding efficiency while maintaining prediction accuracy.
4Adaptability or versatility
If a table of weight coefficients is used for GBI prediction, then prediction versatility is improved, but the table size increases
Solution Approach 1:
The patent optimizes the weight coefficient table by carefully selecting and adjusting the parameter values (weight coefficients) stored in the table. Instead of storing all possible coefficient values, the system uses a compact table with strategically chosen coefficients that cover the necessary prediction scenarios, thereby reducing table size while maintaining prediction versatility.
Data Source
AI summary
A WP prediction and a GBI prediction are adaptively used in combination. A video decoding apparatus (31) includes: a weighted prediction unit (3094) configured to generate a prediction image by using at least one of a weight coefficient or an offset configured for a first unit region and a weight coefficient configured for a second unit region different from the first unit region, the weight coefficient and the offset each being applied to a pixel value included in one or a plurality of reference images.