Picture Encoding and Decoding With Weighted Bi-Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image and video coding schemes face challenges in efficiently leveraging spatial and temporal redundancy due to limitations in bi-prediction and illumination compensation, leading to inefficiencies in compression and reconstruction processes.
Innovation Solution
The proposed method employs a weighted sum of neighboring reconstructed samples to determine bi-prediction and illumination compensation parameters, optimizing integer precision and minimizing errors through least square minimization, thereby enhancing the accuracy and efficiency of picture encoding and decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional bi-prediction and illumination compensation schemes are used, then compression efficiency is achieved, but precision and accuracy in reconstruction are limited
Solution Approach 1:
The patent changes the parameters of bi-prediction by introducing a weighted sum mechanism with configurable weights (w0, w1) and offset (b) parameters. These parameters are optimized through least square minimization to minimize reconstruction error, thereby improving precision while maintaining manageable computational complexity through parameter optimization rather than structural complexity increase
Solution Approach 2:
The patent performs preliminary optimization of the weighted sum parameters (weights and offset) using least square minimization before the actual prediction process. This preliminary action pre-calculates the optimal parameters that minimize reconstruction error, improving reconstruction precision without increasing the complexity of the main prediction operation
2Manufacturing precision
If conventional bi-prediction is used, then compression is achieved, but error minimization and accuracy are insufficient
Solution Approach 1:
The patent implements a feedback mechanism through least square minimization, where the reconstruction error is continuously measured and used to adjust the weighted sum parameters (weights and offset). This iterative feedback process optimizes the parameters to minimize reconstruction error, thereby improving prediction accuracy and reducing information loss
Solution Approach 2:
The patent optimizes the parameters of the weighted sum operation (weights w0, w1 and offset b) by changing them based on least square minimization results. This parameter optimization directly reduces reconstruction error and improves prediction accuracy by finding the parameter set that minimizes the difference between original and reconstructed blocks
3Ease of manufacture
If standard weighted sum operations are used, then computation is simplified, but precision is reduced due to integer constraints
Solution Approach 1:
The patent performs preliminary optimization of the weighted sum parameters using least square minimization to find the optimal weights and offset that minimize reconstruction error. This preliminary action ensures that the integer-weighted sum operation achieves maximum precision within integer constraints, balancing computational simplicity with measurement precision
Solution Approach 2:
The patent changes the parameters of the weighted sum operation (weights and offset) through optimization to achieve the best possible precision within integer arithmetic constraints. By optimizing these parameters rather than using fixed values, the system maintains computational simplicity while maximizing precision
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A method for encoding a picture block is disclosed. A predictor is determined as a weighted sum of sample values of a first motion compensated reference block in a first reference image and of a second motion compensated reference block in a second reference image. Then, a residual block determined from the picture block and the predictor is encoded in a bitstream. Advantageously, the weights of the weighted sum are determined by minimizing an error between reconstructed samples neighboring the picture block and samples neighboring the first and second reference blocks.