Adaptive Transform Kernel Selection for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inter prediction technologies require more bits for compression when there are various movements in blocks, and prediction accuracy is reduced due to the use of a single motion vector for blocks with different objects.
Innovation Solution
A video encoding method that adaptively selects a transform kernel based on the transform shape of a residual block, including square, non-square, or arbitrary shapes, to improve compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single transform kernel is used for all residual blocks, then the device complexity is reduced, but the compression efficiency deteriorates when there are various movements in blocks
Solution Approach 1:
The patent applies dynamics by making the transform kernel selection adaptive rather than fixed. The system dynamically selects transform kernels based on residual block characteristics including motion vectors, prediction modes, and block shapes. This allows the compression system to adjust its behavior according to actual video content, improving compression efficiency while managing complexity through data-driven adaptation.
Solution Approach 2:
The patent changes parameters by varying the transform kernel selection based on multiple factors such as motion vector magnitude, prediction mode (intra/inter), and residual block shape. Different transform kernels are applied depending on these parameter changes, enabling optimized compression for different video regions and conditions without requiring a completely different system for each case.
2Device complexity
If one motion vector is assigned per block, then the device complexity is reduced, but prediction accuracy deteriorates when there are different objects in the same block
Solution Approach 1:
The patent applies segmentation by dividing the block into multiple transform units (TUs) based on residual characteristics. When a block contains multiple objects or complex motion patterns, the system segments the block into smaller TUs that can be processed with appropriate transform kernels. This segmentation allows different motion vectors and prediction modes to be applied to different segments, improving prediction accuracy while managing complexity through hierarchical block structure.
Solution Approach 2:
The patent applies local quality by assigning different transform kernels and prediction parameters to different regions within a block based on their specific characteristics. Regions with different motion vectors, object types, or residual patterns receive tailored processing. This local adaptation improves prediction accuracy for each region while the overall system complexity is managed through standardized processing frameworks.
3Productivity
If adaptive transform kernel selection is implemented, then the compression efficiency is improved, but the device complexity increases due to multiple transform kernels and selection logic
Solution Approach 1:
The patent applies feedback by using residual block characteristics (motion vectors, prediction modes, block shapes) as feedback signals to guide transform kernel selection. The system continuously monitors these characteristics and adjusts the transform kernel choice accordingly. This feedback mechanism enables automatic optimization without manual intervention, improving compression efficiency while the selection logic is driven by objective data metrics.
Solution Approach 2:
The patent applies universality by designing a unified transform kernel selection framework that handles multiple scenarios (intra prediction, inter prediction, different block shapes, various motion patterns) through a single adaptive system. Rather than requiring separate processing paths for each case, the system uses a universal selection mechanism that considers multiple factors and selects appropriate kernels accordingly, reducing overall system complexity while maintaining high compression efficiency.
Data Source
AI summary
Provided is a method of decoding motion information characterized in that information for determining motion-related information includes spatial information and time information, wherein the spatial information indicates a direction of spatial prediction candidates used for sub-units from among spatial prediction candidates located on a left side and an upper side of a current prediction unit, and the time information indicates a reference prediction unit of a previous picture used for prediction of the current prediction unit. Further, an encoding apparatus or a decoding apparatus capable of performing the above described encoding or decoding method may be provided.