Geometric Video Block Transforms for Distorted Intra-Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques suffer from suboptimal compression efficiency due to assumptions of straight lines in video sequences, which are often distorted, and inefficient partitioning of frames into blocks, especially when dealing with non-rectilinear objects, leading to ineffective intra-prediction modes.
Innovation Solution
Applying invertible geometric transformations to video frames or portions before encoding, followed by inverse transformations after reconstruction, to align pixels for improved intra-prediction and partitioning efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If traditional intra-prediction modes are used without geometric transformation, then the encoding process is simple, but compression efficiency is suboptimal due to assumptions of straight lines which are often distorted
Solution Approach 1:
The patent applies geometric transformations (rotation, reflection, scaling) to the current block or reference blocks before performing intra-prediction. This preliminary action transforms distorted or non-rectilinear content into a form that better matches the prediction assumptions, improving compression efficiency without fundamentally changing the prediction algorithm itself
Solution Approach 2:
The patent introduces geometric transformation parameters (rotation angles, reflection axes, scaling factors) that modify the spatial configuration of pixel data. By changing these geometric parameters, the patent adapts the block structure to better represent the actual content geometry, thereby improving prediction accuracy and compression efficiency
2Measurement precision
If frames are partitioned into standard rectangular blocks, then the partitioning process is efficient, but prediction accuracy deteriorates for non-rectilinear objects
Solution Approach 1:
The patent applies geometric transformations to the block partitioning itself or to the reference blocks used for prediction. This preliminary transformation allows the partitioning to better conform to the actual geometry of objects in the video content, improving prediction accuracy for non-rectilinear structures while maintaining efficient processing through standardized transformation operations
Solution Approach 2:
The patent introduces asymmetric transformations (such as selective reflection across different axes or non-uniform scaling) that adapt the rectangular block structure to match asymmetric or distorted object geometries. This allows the prediction process to better handle non-rectilinear objects by transforming them into a more regular configuration
3Measurement precision
If geometric transformation is applied to improve prediction, then intra-prediction accuracy improves, but additional computational steps are required
Solution Approach 1:
The patent performs geometric transformations as a preliminary step before intra-prediction, transforming the current block or reference blocks into a configuration that better matches the prediction model. This preliminary action improves prediction accuracy by aligning the data geometry with the assumptions of the prediction algorithm, while keeping the transformation operations computationally manageable
Solution Approach 2:
The patent uses geometric transformation parameters (rotation angles, reflection axes, scaling factors) to modify the spatial configuration of pixel data. By optimizing these parameters to match the content geometry, the patent improves prediction accuracy while controlling computational complexity through parameter optimization rather than exhaustive search
Data Source
AI summary
Decoding a current block of a current frame includes decoding, from a compressed bitstream, one or more syntax elements indicating that a geometric transformation is to be applied; applying the geometric transformation to at least a portion of the current frame to obtain a transformed portion; and obtaining a prediction of the current block based on the transformed portion and an intra-prediction mode.