Disparity Vector Derivation for 3D-HEVC Inter-View Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face inefficiencies in deriving disparity vectors for inter-view prediction in multi-view video coding, particularly in 3D-HEVC, which is computationally intensive and requires multiple memory accesses, limiting coding efficiency and complexity.
Innovation Solution
The approach involves deriving a disparity vector for a representative block within a group of blocks, rather than individually for each block, and using this vector for inter-view prediction across multiple blocks, reducing the number of neighboring blocks evaluated and eliminating the need for storing IDV flags, thus simplifying the disparity vector derivation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If disparity vectors are derived individually for each block, then prediction accuracy is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent merges the disparity vector derivation process by deriving a single disparity vector for the entire coding tree unit (CTU) rather than individually for each block. This is achieved by evaluating neighboring blocks at the CTU level and using the derived disparity vector for all blocks within that CTU, thereby reducing computational complexity while maintaining prediction accuracy through the unified derivation approach.
Solution Approach 2:
The patent segments the video picture into coding tree units (CTUs) as the fundamental unit for disparity vector derivation. Each CTU is independently processed to derive its disparity vector, which then applies to all blocks within that CTU. This segmentation allows for efficient parallel processing while reducing the overall computational burden compared to block-level derivation.
2Measurement precision
If disparity vectors are derived for each block, then prediction accuracy is improved, but the number of memory accesses increases
Solution Approach 1:
The patent combines multiple memory access operations into a single process by deriving the disparity vector once at the CTU level rather than repeatedly for each block. This reduces the total number of memory accesses to neighboring blocks and IDV flag structures, as the derived disparity vector is reused for all blocks within the CTU, thereby decreasing memory traffic while maintaining prediction accuracy.
3Productivity
If individual disparity vector derivation is performed for each block, then coding efficiency is improved, but processing time increases
Solution Approach 1:
The patent performs the disparity vector derivation as a preliminary action at the CTU level before processing individual blocks. By evaluating neighboring blocks and deriving the disparity vector once for the entire CTU, the system avoids repeated derivation operations for each block, thereby reducing processing time while maintaining coding efficiency through the pre-derived vector's application across all blocks.
Solution Approach 2:
The patent merges multiple derivation operations into a single CTU-level process, which reduces the total processing time. The derived disparity vector is then efficiently applied to all blocks within the CTU, eliminating redundant computations and reducing overall processing time while preserving coding efficiency through the unified derivation approach.
Data Source
AI summary
A parent block is partitioned into the plurality of blocks and a disparity vector derivation process is performed to derive a disparity vector for a representative block in the plurality of blocks. A video encoder generates a bitstream that includes a coded representation of the video data in part by performing, based on the derived disparity vector and without separately deriving disparity vectors for any block in the plurality of blocks other than the representative block, inter-view prediction for two or more blocks in the plurality of blocks. A video decoder reconstructs sample blocks for two or more blocks in the plurality of blocks in part by performing, based on the derived disparity vector and without separately deriving disparity vectors for any block in the plurality of blocks other than the representative block, inter-view prediction for the two or more blocks in the plurality of blocks.