Partial Sub-PU Merge Candidate Construction for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High Efficiency Video Coding (HEVC) faces inefficiencies in motion vector coding due to the generation and management of merge candidate lists, particularly in the computation time required for generating complete sets of motion information for sub-PU merge candidates, which can be redundant and costly in terms of processing resources.
Innovation Solution
The proposed method reduces merge candidate signaling by partially constructing sub-PU merge candidates and performing simplified pruning operations, allowing for the identification and removal of redundant candidates, thereby reducing the computational burden and improving efficiency in generating merge candidate lists.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complete sets of motion information are generated for all sub-PU merge candidates, then coding efficiency is improved, but computation time and processing resources are excessively consumed
Solution Approach 1:
The patent applies partial action by generating motion information only for a subset of sub-PUs within each merge candidate block rather than for all sub-PUs. Specifically, motion information is generated for at most one sub-PU per merge candidate, and only when that candidate is selected for encoding. This partial generation approach maintains coding efficiency by ensuring motion information is available when needed, while dramatically reducing computation time by avoiding redundant generation for unselected candidates.
Solution Approach 2:
The system implements self-service through on-demand completion of merge candidate information. When a merge candidate is selected during encoding, the motion information for its sub-PUs is completed at that moment rather than being pre-computed. This allows the system to service the specific need (encoding a particular block) without the overhead of preparing all possible candidates in advance, thus improving the time efficiency without sacrificing coding quality.
2Measurement precision
If all merge candidates are fully constructed and pruned, then accurate candidate selection is achieved, but processing complexity and time increase significantly
Solution Approach 1:
The patent reduces processing complexity by performing partial construction of merge candidates. Instead of fully constructing all candidates and then pruning, the system partially constructs candidates with minimal motion information initially, performs pruning based on this partial information, and only fully constructs the motion information for candidates that survive the pruning process and are ultimately selected. This approach maintains selection accuracy while significantly reducing the computational complexity of candidate construction and pruning.
Solution Approach 2:
The patent segments the merge candidate processing into distinct phases: initial partial construction, pruning based on partial information, and conditional completion. By dividing the processing workflow into these segments, the system avoids the complexity of fully constructing all candidates upfront. Each segment performs a specific function with reduced complexity, and the segmentation allows the system to eliminate unnecessary candidates early in the process before investing full processing resources.
3Productivity
If redundant merge candidates are removed through pruning, then encoding efficiency is improved, but the pruning process itself consumes additional processing resources
Solution Approach 1:
The patent optimizes the pruning process by using partial motion information rather than complete motion information for candidate comparison. Since full motion information generation is computationally expensive, the system performs pruning based on the partially constructed candidates, which requires fewer processing resources. This approach maintains encoding efficiency by still enabling effective pruning of redundant candidates, while reducing the energy and processing resource consumption associated with the pruning operation itself.
Data Source
AI summary
A reduced merge candidate signaling method is provided. When building a merge candidate list for a prediction unit (PU) of a block of pixels, a video codec skips or partially skips the construction of some sub-PU merge candidates. The video codec then performs simplified pruning operations on the merge candidate list based on the skipped or partially constructed sub-PU merge candidates. The pruned candidate list is then used to select a merge candidate to encode or decode the block of pixels.