Predictive Block Partitioning for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding schemes, such as HEVC, face inefficiencies in block partitioning for arbitrary shaped moving objects, leading to increased coding overhead and complexity due to reliance on rectangular block partitioning, which is not flexible enough to handle object boundaries effectively.
Innovation Solution
The implementation of a predictive block partitioning method using a temporal projection process based on the motion vector field of a reference picture to generate partitioning predictors for block partitioning in the current picture, allowing for more flexible geometric and object-based partitioning methods like segmentation-based partitioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If rectangular block partitioning is used in HEVC, then the block partitioning structure is simple and standardized, but it leads to fine block partitioning along object boundaries and increased coding overhead for arbitrary shaped moving objects
Solution Approach 1:
The patent applies preliminary action by generating a motion trajectory in advance based on reference picture motion vectors before performing block partitioning. This pre-calculated trajectory serves as a guide for determining prediction unit boundaries, allowing the encoder to avoid exhaustive search and reduce coding overhead while maintaining accurate object boundary representation.
Solution Approach 2:
The patent changes the partitioning parameters from fixed rectangular grids to dynamic boundaries defined by motion trajectories. By using the calculated motion trajectory to determine prediction unit shapes and sizes, the system adapts the partitioning parameters to match actual object motion patterns, reducing redundant partitions along object boundaries.
2Adaptability or versatility
If more flexible block partitioning methods are applied, then object boundaries can be handled more effectively, but more side-information is needed and encoder complexity increases significantly
Solution Approach 1:
The patent uses preliminary action by pre-calculating motion trajectories from reference pictures before determining current picture partitioning. This approach provides a ready-made guide for flexible partitioning without requiring exhaustive search, thus achieving adaptability while controlling encoder complexity.
Solution Approach 2:
The patent applies copying by reusing motion vector information from reference pictures to generate motion trajectories for current picture partitioning. Instead of performing complex new calculations, the system copies and adapts existing motion data, reducing computational complexity while maintaining partitioning flexibility.
3Productivity
If rate-distortion optimization is used to determine optimal partitioning, then coding efficiency is improved, but the process requires exhaustive search and is highly specific to multiple conditions increasing complexity
Solution Approach 1:
The patent applies preliminary action by pre-calculating motion trajectories before rate-distortion optimization. This provides a strong initial hint for optimal partitioning, allowing the optimization process to start from a near-optimal state rather than performing exhaustive search from scratch, thus improving coding efficiency while reducing complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention provides a video encoder 100 and a video decoder 200, which may both be used for partitioning a block 301 in a current picture 302 based on at least one partitioning predictor. The encoder 100 and decoder 200 are configured to select at least one reference picture 303 a and a plurality of blocks 304a in the at least one reference picture 303a. Further, to calculate, for each selected block 304a, a projected location in the current picture 302 based on a motion vector associated to the selected block 304a in the reference picture 303a. Then, they are configured to determine each selected block 304a, of which the projected location spatially overlaps with the block 301 in the current picture 302, to be a reference block, and generate for at least one reference block a partitioning predictor based on partitioning information associated to, for example stored in, the at least one reference picture 303 a.