Motion Vector Prediction Using Reliability-Based Candidate Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional motion vector prediction methods suffer from reduced efficiency when there are no effective reference motion vectors nearby, leading to unsuitable predictions and increased bitrate, especially in cases of occlusion or non-rigid objects, where reference motion vectors are not suitable or nonexistent.
Innovation Solution
The method involves selecting a great number of primary candidate blocks to determine primary candidate reference motion vectors, calculating their reliability, narrowing them down to secondary candidate reference motion vectors, and using these to create a predictive motion vector, which is then encoded with the residual, improving prediction efficiency and reducing bitrate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional motion vector prediction methods are used, then encoding process is simple, but prediction efficiency deteriorates when no effective reference motion vectors are available nearby
Solution Approach 1:
The patent segments the motion vector prediction process into multiple stages: first identifying primary candidate blocks adjacent to the current block, then selecting reference motion vectors from these candidates based on reliability criteria. This segmentation allows the system to maintain simplicity for typical cases while providing enhanced prediction capability when needed, resolving the contradiction between prediction efficiency and computational complexity.
Solution Approach 2:
The patent performs preliminary identification of candidate blocks and pre-selection of reference motion vectors before the actual prediction process. By preparing the candidate set in advance based on spatial adjacency and reliability metrics, the system ensures high prediction efficiency is achieved when needed without requiring complex real-time computations during encoding.
2Reliability
If reference motion vectors from adjacent blocks are used, then encoding complexity is low, but prediction accuracy decreases in cases of occlusion or non-rigid objects
Solution Approach 1:
The patent applies local quality by evaluating the reliability of each candidate block individually based on its specific characteristics (such as motion consistency and occlusion detection). Instead of uniformly treating all adjacent blocks, the system selectively chooses reference motion vectors from blocks that locally satisfy reliability criteria, thereby improving prediction accuracy for occluded or non-rigid regions while maintaining encoding simplicity for standard cases.
Solution Approach 2:
The patent changes the selection parameter from simple spatial adjacency to a composite reliability metric that considers motion consistency, block type, and occlusion probability. This parameter change enables the system to distinguish between reliable and unreliable reference blocks, improving prediction accuracy without significantly increasing encoding complexity through efficient metric calculation.
3Reliability
If more reference motion vectors are considered, then prediction efficiency improves, but encoding time and computational load increase
Solution Approach 1:
The patent applies partial action by considering only a limited set of primary candidate blocks (typically four adjacent blocks) rather than all possible blocks in the picture. This partial evaluation provides sufficient prediction efficiency for most cases while avoiding the excessive computational load of examining all blocks, thus maintaining encoding speed.
Solution Approach 2:
By pre-identifying and prioritizing the most likely candidate blocks (those with highest reliability scores) before performing detailed prediction, the system achieves high prediction efficiency with minimal computational effort. The preliminary filtering step ensures that only the most promising candidates are processed further, maintaining encoding speed while improving accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In motion vector prediction encoding, the efficiency of motion vector prediction is improved and the efficiency of moving picture encoding is improved. A motion search is performed for a block to be encoded in a picture to be encoded using an encoded reference picture to calculate a motion vector. A plurality of blocks which include at least one of encoded blocks in the picture to be encoded and encoded blocks in an encoded picture and in predetermined positions relative to the position of the block to be encoded in the picture to be encoded are set as primary candidate blocks, and N primary candidate reference motion vectors are determined from motion vectors used in encoding the primary candidate blocks. The degrees of reliability of the primary candidate reference motion vectors which quantitatively represent effectiveness in motion vector prediction of the block to be encoded is calculated using encoded picture information for each of the primary candidate reference motion vectors. Top M primary candidate reference motion vectors with higher degrees of reliability are selected from the primary candidate reference motion vectors as secondary candidate reference motion vectors. A predictive motion vector of the block to be encoded is calculated using the secondary candidate reference motion vectors, and a residual between the motion vector obtained through the motion search of the block to be encoded and the predictive motion vector is encoded as encoded information of the motion vector.