Prediction-Derived Blending for Geometric Video Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies face inefficiencies in geometric partitioning modes due to block dependencies and the need for reconstructed neighboring blocks to derive blending function parameters, which can hinder performance.
Innovation Solution
Derive blending function parameters for geometric partitioning modes using predictions (P0 and P1) without relying on reconstructed template samples, allowing for spatially varying bi-predictions based on regression methods and parametric functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reconstructed neighboring blocks are used to derive blending function parameters in geometric partitioning modes, then prediction accuracy is improved, but block dependencies increase and processing complexity worsens
Solution Approach 1:
The patent segments the block into multiple partitions (e.g., triangular or rectangular partitions) and derives blending function parameters independently for each partition using only locally available prediction samples. This eliminates the need to wait for reconstructed neighboring blocks while maintaining prediction accuracy through partition-specific parameter derivation.
Solution Approach 2:
The patent performs preliminary derivation of blending function parameters using prediction samples that are already available before reconstruction of neighboring blocks is complete. By using prediction samples from the current block and previously decoded blocks, the method prepares the blending parameters in advance, avoiding delays caused by block dependencies.
2Measurement precision
If reconstructed neighboring blocks are used to derive blending function parameters, then prediction accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary derivation of blending function parameters using prediction samples that are already available before reconstruction of neighboring blocks is complete. By using prediction samples from the current block and previously decoded blocks, the method prepares the blending parameters in advance, avoiding delays caused by block dependencies.
Solution Approach 2:
The patent uses temporary prediction samples and intermediate calculation results that are discarded after parameter derivation. Instead of storing and processing full reconstructed blocks, the method uses lightweight prediction samples and computes blending parameters on-the-fly, reducing memory access time and processing overhead.
3Productivity
If spatially varying bi-predictions with regression methods are implemented, then compression efficiency is improved, but computational complexity increases
Solution Approach 1:
The patent changes the parameters of the blending function (e.g., weight coefficients, partition boundaries) based on local characteristics of each partition. By adapting these parameters to match local prediction errors and sample variations, the method achieves better compression efficiency while keeping the functional form simple and computationally manageable.
Solution Approach 2:
The patent applies different blending function parameters to different spatial partitions within the block. Each partition receives customized parameters derived from its local prediction samples, allowing the system to capture local variations in motion and texture while avoiding the need for complex global optimization.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In various implementations, method and devices are disclosed that encode or decode a block of a picture with a geometric partition that is responsive to a blending parametric function wherein the parameters of the blending parametric function are derived from samples of at least one of the first prediction or the second prediction; and wherein the sample-wise weighted blending of the first prediction and the second prediction is determined with a weight output from the blending parametric function responsive to the derived parameters applied to a sample location in the block.