Transform Basis Symmetry for Prediction Residual Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding schemes face inefficiencies in reconstructing prediction residuals for transform units and complexity in adaptive core transforms due to the number of transform bases.
Innovation Solution
An image decoding and coding apparatus that performs inverse transforms using a selected transform basis, which includes a first and a second transform basis obtained by symmetric transformation of the first basis, allowing for efficient reconstruction of prediction residuals and reducing complexity by sharing transform matrices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple transform bases are used for adaptive core transform, then the accuracy of prediction residual reconstruction is improved, but the device complexity increases
Solution Approach 1:
The patent merges the storage and processing of multiple transform bases by recognizing that symmetric transform bases can be generated from a single base matrix through symmetry operations. Instead of storing separate matrices for each transform basis, the system stores one base matrix and generates others on-demand using symmetry relationships, thereby reducing storage requirements and simplifying the processing architecture while maintaining access to multiple transform bases for accurate prediction residual reconstruction.
Solution Approach 2:
The patent makes a single transform basis matrix serve multiple functions by using it to generate multiple symmetric transform bases. The same base matrix is used to create different transform variations through symmetry operations (horizontal, vertical, and diagonal symmetries), allowing one matrix to fulfill the role of multiple transform bases. This multi-functionality reduces the overall system complexity while preserving the ability to select from multiple transform options for optimal reconstruction accuracy.
2Adaptability or versatility
If multiple transform bases are stored, then the versatility of transform selection is improved, but the memory requirements increase
Solution Approach 1:
The patent extracts the essential information from multiple transform bases by identifying and storing only the fundamental base matrix. The redundant information contained in symmetric variations of the transform bases is removed, as these can be regenerated from the single stored matrix through symmetry operations. This extraction process retains the versatility of having multiple transform options available while dramatically reducing the memory storage requirements by eliminating redundant data.
Solution Approach 2:
The patent creates copies of the transform basis through symmetry operations rather than storing independent copies. Instead of storing multiple full transform matrices in memory, the system stores one base matrix and generates symmetric variations (horizontal flip, vertical flip, diagonal flip) as needed during processing. This copying approach maintains transform selection flexibility while using minimal memory, as the "copies" are generated computationally rather than stored physically.
Data Source
AI summary
An image decoding apparatus and an image coding apparatus are implemented that are capable of more efficiently and effectively performing reconstruction of a prediction residual for a TU. An image decoding apparatus (1) is an image decoding apparatus including: a transforming unit (15) configured to perform, on a prescribed unit basis, inverse transform of a transform coefficient or a modification transform coefficient by using a transform basis selected from a plurality of the transform bases, in which the plurality of the transform bases include a first transform basis and a second transform basis obtained by performing symmetric transform of the first transform basis.