Motion Vector Prediction Pruning for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video coding systems face challenges in parsing bitstreams when the reference frame specifying the temporally co-located motion vector predictor (MVP) is lost, leading to inefficiencies in motion compensation and coding efficiency losses due to pruning processes.
Innovation Solution
The proposed solution involves forming an intermediate list of spatial MVPs, pruning redundant spatial MVPs, and then combining the temporally co-located MVP with the pruned list to maintain coding efficiency while ensuring the bitstream can still be parsed, even if the reference frame is lost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the reference frame specifying the temporally co-located motion vector predictor is lost, then parsing failures occur in the bitstream, but maintaining the conventional MVP list structure prevents efficient use of spatial MVPs
Solution Approach 1:
The patent segments the MVP list construction into two independent phases: first constructing a pruned list of spatial MVPs, then separately adding the temporal MVP if available. This segmentation allows the spatial MVPs to be reliably processed even when the temporal reference frame is lost, while still enabling efficient coding when the frame is available.
Solution Approach 2:
The patent performs preliminary pruning of spatial MVPs before adding the temporal MVP to the list. This preliminary action ensures that the spatial MVPs are optimized for coding efficiency independent of the temporal reference frame status, allowing the system to maintain high coding efficiency while being robust to reference frame losses.
2Productivity
If spatial MVPs are pruned to remove redundancy, then coding efficiency improves, but parsing failures occur when the temporal reference frame is lost
Solution Approach 1:
The patent separates the pruning operation to apply only to spatial MVPs, excluding the temporal MVP from this process. This creates an intermediate pruned list that maintains coding efficiency while preserving the temporal MVP as a separate, reliable element that ensures parsing success even when spatial references are unavailable.
Solution Approach 2:
The patent introduces an intermediate pruned list of spatial MVPs as a mediator between the full spatial MVP list and the final MVP list. This intermediate structure allows aggressive pruning for efficiency while the temporal MVP acts as a fallback mediator that ensures parsing reliability when spatial references are lost.
3Measurement precision
If motion vector search is performed instead of using MVPs, then motion compensation accuracy improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary selection of candidate motion vectors from spatial and temporal neighbors before the actual motion compensation process. This preliminary action creates an optimized MVP list that provides accurate motion estimation without requiring exhaustive search, thereby maintaining high accuracy while reducing computational complexity.
Solution Approach 2:
The patent enables the MVP list to serve itself by automatically selecting and pruning candidate motion vectors from available spatial and temporal references. This self-service mechanism provides accurate motion compensation predictions without external intervention or complex search algorithms, balancing accuracy and computational efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In general, techniques are described for performing motion vector prediction for video coding. An apparatus comprising a motion compensation unit may implement the techniques. The motion compensation unit determines spatial candidate motion vectors (MVPs) associated with a current portion of a video frame and prunes the spatial candidate motion vectors to remove duplicates without removing a temporal candidate motion vector. The motion compensation unit selects one of the temporal candidate motion vector or one of the spatial candidate motion vectors remaining after pruning as a selected candidate motion vector based on a motion vector predictor (MVP) index signaled in a bitstream and performs motion compensation based on the selected candidate motion vector.