Subband Matching Pursuit Coding With Band-Adaptive Dictionaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data compression techniques for multi-dimensional digital data, such as images and audio, face challenges in reducing computational complexity while maintaining data integrity and accuracy, especially in higher dimensions, due to the large size of codebooks and the complexity of calculations involved in matching pursuits processes.
Innovation Solution
The implementation of a reduced dictionary using a mask to select a subset of a full n-dimensional dictionary, allowing for separable basis functions to be applied in different dimensions, which reduces the number of calculations and complexity, and the use of different codebooks for different regions of transformed data, such as those produced by wavelet decomposition, to improve PSNR and reduce bit cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full n-dimensional dictionary is used for matching pursuits coding, then data integrity and accuracy are maintained, but computational complexity and calculation time increase significantly
Solution Approach 1:
The full n-dimensional dictionary is segmented into multiple sub-dictionaries based on wavelet sub-bands (LL, LH, HL, HH). Each sub-dictionary contains basis functions tailored to specific frequency characteristics of that sub-band, allowing the coding process to operate on smaller, more manageable subsets rather than the complete dictionary at once.
Solution Approach 2:
Different sub-dictionaries are designed with locally optimized basis functions that match the statistical properties and frequency content of specific wavelet sub-bands. For example, low-pass sub-bands use different basis functions compared to high-pass sub-bands, providing locally adapted representation that maintains accuracy while reducing overall complexity.
2Measurement precision
If a full codebook is used for matching pursuits coding, then coding accuracy is maintained, but bit cost increases
Solution Approach 1:
The codebook is segmented into multiple sub-codebooks corresponding to different wavelet sub-bands. Each sub-codebook is optimized for its specific sub-band characteristics, allowing more efficient representation of coefficients in that band. This segmentation enables selective coding where only relevant sub-codebooks are activated for each block, reducing overall bit cost.
Solution Approach 2:
The invention changes the parameter of codebook selection by adapting the active sub-codebooks based on the energy distribution and characteristics of different wavelet sub-bands. This dynamic parameter adjustment allows the system to use smaller, more efficient codebooks for low-energy sub-bands while maintaining larger codebooks only where necessary, optimizing the rate-distortion tradeoff.
3Measurement precision
If matching pursuits processes are applied to higher dimensional data, then data representation accuracy improves, but computational feasibility deteriorates
Solution Approach 1:
Higher dimensional data processing is achieved by segmenting the problem into multiple lower-dimensional sub-problems through wavelet decomposition. The matching pursuits coding is applied separately to each wavelet sub-band, which has reduced dimensionality compared to the original full-resolution data. This segmentation makes high-dimensional coding computationally feasible while maintaining overall accuracy through the combination of sub-band representations.
Solution Approach 2:
The invention transforms the high-dimensional coding problem into a multi-dimensional solution space by applying wavelet decomposition across different frequency dimensions. This creates a hierarchical structure where coding is performed at multiple resolution levels and frequency bands, converting an intractable high-dimensional problem into a series of manageable lower-dimensional sub-problems that can be solved efficiently.
Data Source
AI summary
A method, system, and computer program product are used for coding multi-dimensional data using a codebook of basis functions. A transform is applied to the multi-dimensional data to create transform coefficients. Groups of transform coefficients are formed. Different groups of transform coefficients are coded with non-identical dictionaries of basis functions from the codebook.