Affine Inter-Prediction with Translated Reference Areas for Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of uniform square coding units in high-resolution images leads to quality degradation in reconstructed images during video compression, necessitating improved methods for splitting coding units into various shapes.
Innovation Solution
A method for video decoding that involves determining a central motion vector, adjusting a reference area by parallel translation, and using affine model-based inter-prediction to handle motion vectors deviating from picture boundaries, allowing for non-square coding units and improved image reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If uniform square coding units are used for high-resolution images, then the coding process is simple, but the quality of reconstructed images degrades
Solution Approach 1:
The current block is divided into multiple sub-blocks, and each sub-block is processed independently with its own motion compensation. This segmentation allows for more precise local modeling of motion patterns while maintaining overall coding efficiency.
Solution Approach 2:
Different motion models are applied to different sub-blocks based on their local characteristics. The affine motion model is used for sub-blocks where it provides better prediction accuracy, while other models can be used elsewhere, achieving local optimization of prediction quality.
2Measurement precision
If affine model-based inter-prediction is performed with motion vectors pointing outside reference picture boundaries, then motion prediction accuracy improves, but reference area access becomes problematic
Solution Approach 1:
Instead of clipping the motion vector to stay within boundaries, the patent inverts the approach by allowing the motion vector to point outside boundaries and then translating the reference area back into the current picture. This preserves the accuracy of motion vectors while ensuring reliable reference access.
Solution Approach 2:
The patent transforms the problem from a 2D reference picture coordinate system to a 1D array representation. By mapping 2D coordinates (x, y) to 1D array indices and applying translation in this transformed space, the reference area can be reliably accessed even when motion vectors extend beyond picture boundaries.
Data Source
AI summary
A video decoding method includes, determining a center motion vector of a current block by using a base motion vector of the current block based on affine model-based inter-prediction being performed in the current block, determining a reference range of an area to be referred to, with respect to the current block, based on a size of the current block, based on a reference area having a size of the reference range with respect to a point in a reference picture of the current block, the point being indicated by a central motion vector of the current block, deviating from or including a boundary of the reference picture, changing the reference area by parallelly translating the reference area into a current picture, and determining prediction samples of sub-blocks of the current block in the changed reference area from the reference picture.