Transform Matrix Determination for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital signal transformation methods in video and image coding, such as those in H.264/AVC and AVS, face inefficiencies due to insufficient decorrelation capability, excessive transform distortion, and unsatisfactory normalization of transform kernels, particularly when dealing with large transform sizes like 32×32 and 64×64, which complicates calculations and affects bit-width management.
Innovation Solution
A method and device for determining a transform matrix by calculating minimum and maximum value matrices and a threshold value based on DCT matrices, ensuring elements fall within specific ranges and satisfy conditions related to transform distortion, normalization, and orthogonality, thereby improving the transformation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional transform methods (H.264/AVC, AVS) are used for large transform sizes (32×32, 64×64), then transformation can be performed, but transform distortion increases and decorrelation capability becomes insufficient
Solution Approach 1:
The patent modifies the transform kernel parameters by introducing a normalization factor and adjusting the transform matrix elements. Specifically, the transform kernel is normalized by dividing by the square root of the transform size (N), and the matrix elements are scaled by a factor (e.g., 1/8 or 1/16) to control the magnitude of transform coefficients. This parameter adjustment reduces transform distortion while maintaining decorrelation capability for large transform sizes.
2Manufacturing precision
If transform size is increased to 32×32 or 64×64, then better frequency transformation is achieved, but calculation complexity increases and bit-width management becomes difficult
Solution Approach 1:
The patent introduces scaling parameters (e.g., dividing transform coefficients by 8 or 16) to control the bit-width of intermediate calculation results. By adjusting these scaling factors, the patent manages the complexity of calculations for large transform sizes while maintaining transformation performance. The normalized transform kernel also helps in controlling the dynamic range of transform coefficients.
3Measurement precision
If transform kernel normalization is improved, then transformation accuracy increases, but device complexity increases due to additional calculations
Solution Approach 1:
The patent pre-calculates and stores the normalized transform kernel values in lookup tables during initialization. Instead of performing complex normalization calculations during each transformation operation, the patent uses these pre-computed values directly. This approach maintains high normalization accuracy while significantly reducing the computational overhead during actual transform operations.
Data Source
AI summary
Provided are a signal transforming method and a signal transforming device. For example, the signal transforming method includes determining a minimum-value matrix and a maximum-value matrix with respect to elements of a matrix used in frequency transformation, wherein the minimum-value matrix is configured of elements of minimum value and the maximum-value matrix is configured of elements of maximum value; determining a maximum threshold value of a result value of a function indicating at least one selected from transform distortion, normalization, and orthogonality of the matrix; determining a transform matrix configured of elements that are greater than the elements of the minimum-value matrix and less than the elements of the maximum-value matrix at respective positions of the matrix, and in which the result value of the function is less than the maximum threshold value; and transforming an input signal by using the determined transform matrix.