Visual Error Weights for JPEG2000 Rate Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression techniques, such as JPEG 2000, rely on mean squared error (MSE) for rate allocation, which fails to accurately represent visual quality, leading to artifacts in reconstructed images and increased computational complexity in attempts to improve visual masking.
Innovation Solution
Generating visual error weights based on second-order or higher moments and average absolute values of wavelet coefficient values, allowing for per-codeblock or per-segment bit allocation that prioritizes less busy image areas, thereby hiding errors in busy regions and reducing errors in consistent regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If mean squared error (MSE) is used for rate allocation in JPEG 2000 encoding, then the encoding process is simple and computationally efficient, but the visual quality of reconstructed images deteriorates due to artifacts in important regions
Solution Approach 1:
The patent applies local quality by computing visual error weights for each codeblock based on its position and characteristics in the image. Different codeblocks receive different weighting factors that reflect their visual importance, allowing the encoder to allocate bits locally according to visual significance rather than uniformly based on MSE. This resolves the contradiction by maintaining simple global MSE computation while adding local differentiation through weight factors.
Solution Approach 2:
The patent changes the error metric parameter from plain MSE to weighted MSE by introducing visual error weights. These weights are computed based on codeblock characteristics (position, frequency sub-band, etc.) and modify the traditional MSE calculation. This parameter change allows the encoding to prioritize visually important regions without fundamentally changing the MSE-based rate allocation framework, thus improving visual quality while maintaining computational efficiency.
2Manufacturing precision
If fractional moments are computed to improve visual masking in rate allocation, then visual quality improves, but computational complexity increases significantly
Solution Approach 1:
The patent uses integer order moments (first, second, third moments) as computationally inexpensive alternatives to fractional moments. These integer moments can be computed efficiently using simple arithmetic operations on wavelet coefficients, providing the necessary visual masking information without the high computational cost of fractional moment calculations. The patent effectively replaces expensive computational objects with cheaper, sufficient alternatives.
Solution Approach 2:
The patent changes the mathematical parameter from fractional moments to integer order moments. This substitution maintains the ability to capture visual masking effects while dramatically reducing computational complexity. Integer moments involve only integer power calculations and averaging, whereas fractional moments require complex root calculations and iterative methods, thus resolving the contradiction between visual quality improvement and computational burden.
3Measurement precision
If more bits are allocated to all codeblocks to reduce MSE, then the mean squared error decreases, but the visual quality does not improve proportionally due to masking effects in different regions
Solution Approach 1:
The patent applies local quality by differentiating bit allocation across codeblocks based on their visual importance. Instead of uniformly reducing MSE across all regions, the patent identifies visually sensitive regions (using visual error weights) and allocates more bits specifically to those regions. This local differentiation ensures that MSE reduction efforts are concentrated where they will have the greatest visual impact, resolving the contradiction between MSE reduction and visual quality improvement.
Solution Approach 2:
The patent changes the error metric from unweighted MSE to weighted MSE, where the weights reflect visual masking characteristics. This parameter change transforms the optimization goal from minimizing raw pixel differences to minimizing perceptually significant errors. By weighting the MSE calculation, the patent ensures that bit allocation decisions are driven by visual quality metrics rather than raw numerical error metrics.
Data Source
AI summary
A low complexity visual masking method used as part of an image encoding process is described. The method is suitable for use in JPEG2000 image compression systems. Control weights used for rate allocation are generated based on integer order moments of wavelet transformed coefficients corresponding to a codeblock. The novel rate allocation weight generation method can, and in some embodiments is, combined with an apriori rate allocation algorithm, where allocation of bits to different portions of images is controlled as a function of one or more generated weights. The methods and apparatus of the present invention have the effect of increasing errors in busy areas of an image where they tend to be less noticeable and allocating a higher number of bits to less busy areas than some other systems, e.g., systems which attempt to minimize a mean squared error under a constraint of a user selected output rate.