Affine Prediction Using Epipolar Curves for Resolution Mismatch
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex motion inherent in video from moving cameras, and/or cameras with different resolution or distortion profiles, increases the need for accurate, flexible prediction modeling for compression, but sophisticated motion models have high complexity in parameter determination.
Innovation Solution
Using epipolar-based constraints to reduce the complexity of estimating affine predictors by determining an epipolar curve from corners of the block, based on known camera geometry, and constraining the search to a single parameter along this curve for affine prediction between pictures with different resolutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated motion models are used for accurate prediction, then prediction accuracy is improved, but parameter determination complexity increases
Solution Approach 1:
The patent introduces epipolar curves as an intermediary constraint to bridge the current picture and reference picture. These curves, derived from camera geometry, serve as a mediator that guides the search for corresponding blocks, reducing the degrees of freedom in the motion model while maintaining prediction accuracy for complex motions.
Solution Approach 2:
The patent changes the parameter space by constraining the affine motion model to search only along epipolar curves rather than across the entire reference picture. This parameter constraint reduces the complexity of determining motion parameters while preserving the ability to model complex motions through the geometric properties of the epipolar constraint.
2Adaptability or versatility
If affine prediction with multiple parameters is used, then motion modeling flexibility is improved, but encoding complexity increases
Solution Approach 1:
The patent segments the search space into one-dimensional epipolar curves rather than searching the entire two-dimensional reference picture. This segmentation of the search domain maintains motion modeling flexibility by allowing different affine parameters for different regions while reducing encoding complexity through the geometric constraint that limits the search to specific curves.
Solution Approach 2:
The patent transforms the two-dimensional motion search problem into a one-dimensional search along epipolar curves. By introducing the epipolar geometric constraint, the invention reduces the dimensionality of the search space from 2D to 1D, thereby reducing encoding complexity while preserving motion modeling capability through the curve-based search path.
3Measurement precision
If comprehensive block matching is performed, then prediction accuracy is improved, but encoding time increases
Solution Approach 1:
The patent performs preliminary action by pre-defining epipolar curves based on camera geometry before the actual block matching process. This preliminary geometric constraint establishes the search paths in advance, allowing the encoder to quickly locate corresponding blocks along these predetermined curves without performing exhaustive searches, thus reducing encoding time while maintaining prediction accuracy.
Solution Approach 2:
The patent implements skipping by restricting the block matching search to only those regions along epipolar curves that are geometrically plausible. This allows the encoder to skip large portions of the reference picture that cannot contain the true match, rushing through the encoding process by focusing computational effort only on relevant search regions defined by the epipolar constraint.
Data Source
AI summary
An apparatus configured to encode video data is configured to receive a first picture having a first resolution, receive a second picture having a second resolution, wherein the second resolution is different than the first resolution, and wherein the first picture and the second picture have at least partially overlapping fields of view, and encode a current block of the first picture, relative to the second picture, using affine prediction with a single parameter search, wherein the single parameter search is performed along an epipolar curve, and wherein the single parameter search is used to determine parameters of an affine prediction model.