Affine Inter Prediction Using Control Point Motion Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-resolution and high-quality image data requires efficient compression techniques to reduce transmission and storage costs, as conventional methods struggle with the increased data volume, especially when images undergo transformations like rotation, zooming, and deformation.
Innovation Solution
An affine motion model-based inter prediction method that derives sample unit motion vectors from control points, allowing for efficient inter prediction by using motion vectors of neighboring blocks, reducing data requirements and improving coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If conventional inter prediction methods are used for high-resolution images, then transmission and storage costs increase due to increased data volume, but implementing advanced compression techniques increases device complexity
Solution Approach 1:
The current block is divided into multiple sample units, and motion vectors are derived for each sample unit independently based on control points. This segmentation allows for more precise motion compensation while maintaining computational efficiency by processing smaller units rather than the entire block at once.
Solution Approach 2:
The patent changes the coordinate system origin to the bottom-right sample (0,0) and defines control point coordinates relative to this origin. This parameter change simplifies the affine transformation calculations and enables more efficient derivation of sample unit motion vectors from control point motion vectors.
2Measurement precision
If affine motion model with control points is implemented, then inter prediction accuracy improves for transformed images, but data requirements for motion vectors increase
Solution Approach 1:
The patent extracts only the essential control point motion vectors from the full affine motion model and uses these to derive sample unit motion vectors. By taking out and utilizing only the necessary control point information rather than transmitting complete motion field data, the method reduces data requirements while maintaining prediction accuracy for transformed images.
Solution Approach 2:
Instead of applying full affine transformation to all samples, the method applies partial action by deriving motion vectors only for control points and then interpolating to sample units. This partial approach provides sufficient accuracy for most practical cases while significantly reducing computational and data requirements compared to complete affine modeling.
3Productivity
If sample unit motion vectors are derived from control points, then coding efficiency improves, but device complexity increases due to additional derivation steps
Solution Approach 1:
The patent performs preliminary action by pre-defining control point positions and coordinates relative to the bottom-right origin before deriving sample unit motion vectors. This preliminary setup establishes a consistent coordinate framework that simplifies subsequent motion vector derivations and reduces the complexity of the overall process.
Solution Approach 2:
Control points serve as intermediaries between the reference picture and sample units. Instead of directly computing motion vectors for all sample units, the method uses control points as intermediate reference points, which simplifies the derivation process and improves coding efficiency by reducing the number of direct computations required.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to the present invention, a video decoding method performed by a decoding device comprises the steps of: deriving control points (CPs) for a current block; acquiring motion vectors for the CPs; deriving a sample-unit motion vector in the current block on the basis of the acquired motion vectors; deriving a prediction sample for the current block on the basis of the sample-unit motion vector; and generating a restoration sample on the basis of the prediction sample, wherein coordinates of a bottom-right sample position of the current block are (0, 0), and coordinates of CP0 among the CPs are (0, 0). According to the present invention, inter prediction can be effectively performed through sample-unit motion vectors even when an image in a current block is planarly moved and when the image is variously changed, and since motion vectors for CPs can be derived on the basis of already decoded neighboring blocks, a data amount for information on the motion vectors for the CPs and for a residual signal for the current block can be removed or reduced, and overall coding efficiency can be improved.