Inter-view Residual Prediction in 3D Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced residual prediction techniques in 3D-HEVC face challenges such as increased cache miss rates and memory bandwidth requirements due to the need to access multiple reference pictures for different prediction units within a picture, leading to slower coding processes and higher computational complexity.
Innovation Solution
Scaling motion vectors of prediction units towards a fixed reference picture within the same view, reducing the need to access multiple reference pictures and using a bi-linear filter for motion-compensated block generation to minimize memory access and computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple reference pictures are accessed for different prediction units within a picture, then prediction accuracy is improved, but cache miss rates and memory bandwidth requirements increase
Solution Approach 1:
The patent merges the reference picture selection process by determining a single fixed reference picture for all prediction units within a picture. This consolidation reduces the number of memory accesses to multiple reference pictures, thereby lowering cache miss rates and memory bandwidth requirements while maintaining prediction accuracy through the unified reference framework.
Solution Approach 2:
The fixed reference picture serves multiple functions across different prediction units within the same picture. Instead of each prediction unit requiring its own reference picture, the single fixed reference picture is universally used for all PUs, reducing memory access overhead and improving coding efficiency without sacrificing prediction quality.
2Measurement precision
If multiple reference pictures are accessed for different prediction units within a picture, then prediction accuracy is improved, but coding speed deteriorates
Solution Approach 1:
The patent merges the reference picture selection process by determining a single fixed reference picture for all prediction units within a picture. This consolidation reduces the number of memory accesses to multiple reference pictures, thereby lowering cache miss rates and memory bandwidth requirements while maintaining prediction accuracy through the unified reference framework.
3Loss of energy
If motion vectors are scaled towards a fixed reference picture, then memory access is reduced, but motion compensation accuracy may be affected
Solution Approach 1:
The patent applies parameter changes by scaling motion vectors to reference a fixed picture instead of their original reference pictures. This parameter transformation reduces memory access requirements while the scaling process is designed to maintain motion compensation accuracy by preserving the relative spatial relationships in the scaled vectors.
Data Source
AI summary
A video coder scales a motion vector of a current prediction unit (PU) of a current picture in order to compensate for a difference in temporal distance. In addition, the video coder determines a predictive block for the current PU, determines a disparity reference block based on samples of the disparity reference picture at a location indicated by a disparity vector of the current PU, and determines, based on samples of the fixed reference picture at a location indicated by the scaled motion vector, a temporal-disparity reference block for the current PU. The video coder then determines a residual predictor for the current PU. Each sample of the residual predictor for the current PU indicates a difference between a sample of the temporal-disparity reference block for the current PU and a corresponding sample of the disparity reference block.