Image Coding Phase Correlation Reduces Computational Cost
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image coding methods face high computational costs due to the need to test a large number of atoms during atomic decomposition, especially when using iterative methods like matching pursuit, as they require scalar products with a large dictionary that includes both analytical and textured atoms.
Innovation Solution
The method employs a second dictionary with patches that allow for phase correlation to test numerous re-phased atoms efficiently, reducing the number of iterations and computational cost by selecting the most correlated atoms, rather than increasing the dictionary size to include re-phased atoms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the dictionary size is increased by adding re-phased atoms for each textured atom, then the prediction accuracy is improved, but the calculation cost increases significantly
Solution Approach 1:
The patent segments the testing process into two stages: first testing a reduced set of representative atoms from the dictionary, then using phase correlation to efficiently evaluate re-phased variations of selected atoms. This segmentation avoids testing all possible re-phased atoms directly, reducing computational complexity while maintaining prediction accuracy.
Solution Approach 2:
The patent introduces phase correlation as an intermediary mechanism to evaluate the match between atoms and residue vectors. Instead of directly computing scalar products for all re-phased atoms, phase correlation serves as an efficient mediator to identify promising candidates, significantly reducing the number of full scalar product computations required.
2Measurement precision
If the number of atoms tested during atomic decomposition is increased, then the prediction fidelity is improved, but the number of iterations and processing time increase
Solution Approach 1:
The patent performs preliminary testing with a reduced dictionary of representative atoms before proceeding to detailed evaluation with re-phased atoms. This preliminary action identifies promising candidates early, allowing the algorithm to focus computational resources on fewer high-potential atoms, thereby improving processing speed without sacrificing prediction fidelity.
Solution Approach 2:
The patent replaces the traditional mechanical approach of directly computing scalar products for all atoms with a two-stage process involving phase correlation followed by selective scalar product computation. This substitution of the computational mechanism dramatically reduces processing time while maintaining the ability to identify the most correlated atoms.
3Measurement precision
If a large dictionary including both analytical and textured atoms is used, then the decomposition accuracy is improved, but the scalar product computations become excessively costly
Solution Approach 1:
The patent extracts and tests only the most relevant atoms from the large dictionary using phase correlation as a filtering mechanism. By taking out only the promising candidates for further detailed evaluation, the method maintains decomposition accuracy while dramatically reducing the number of energy-intensive scalar product computations required.
Data Source
AI summary
A method for coding a current block is disclosed. The method comprises an iterative step of decomposition into atoms of a first dictionary of a vector of data comprising at least reconstructed image data associated with pixels situated in a causal neighboring area of a current block.At each iteration, the step of decomposition into atoms comprises the following steps for:selecting a first atom in the first dictionary of atoms,applying a phase correlation between a patch associated with the first atom and a current residue vector, when the first atom is a textured atom,extracting from the patch at least the part most correlated with the residue vector, the extracted part forming a second atom,selecting from among the first atom and the second atom, the atom most correlated with the current residue vector, andupdating the residue vector according to the selected atom.