Vector-Valued Image Denoising via Frequency-Specific Covariance Splitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for denoising vector-valued images in iterative image reconstruction, particularly in multi-energy X-ray computed tomography, suffer from crosstalk between material images due to correlated noise models, leading to false diagnostic results.
Innovation Solution
A device and method that modify the covariance matrix by splitting it into frequency-specific submatrices to create separate noise models for high and low spatial frequency bands, allowing for tuned trade-offs between cross-talk reduction and correlated noise removal, using a processor to generate a final loss function for denoising with a noise-suppressor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a correlated noise model with covariance matrix Ci-1 is used for denoising, then correlated noise portions are removed efficiently, but crosstalk between denoised material images occurs
Solution Approach 1:
The patent segments the denoising process into frequency-specific operations by splitting the covariance matrix into frequency submatrices. Different frequency bands (low-frequency and high-frequency) are processed with different correlation assumptions, allowing correlated noise removal in low-frequency regions while preventing crosstalk-induced artifacts in high-frequency regions.
Solution Approach 2:
The patent applies different noise correlation models to different spatial frequency regions of the image. Low-frequency components use a correlated noise model with full covariance matrices, while high-frequency components use an uncorrelated noise model with diagonal covariance matrices, optimizing denoising performance for each region's characteristics.
2Measurement precision
If frequency-specific covariance matrices are used for denoising, then crosstalk is reduced, but processing complexity increases
Solution Approach 1:
The patent divides the covariance matrix into frequency-specific submatrices that can be pre-computed and stored. During denoising, only the relevant frequency submatrices are applied, reducing real-time computational complexity while maintaining the benefits of frequency-specific processing.
Solution Approach 2:
The patent performs frequency decomposition and covariance matrix splitting in advance before the actual denoising operation. This preliminary processing allows the main denoising loop to use pre-prepared frequency-specific covariance matrices, reducing computational burden during iterative reconstruction.
Data Source
AI summary
The present invention relates to a device (100) for denoising a vector-valued image, the device (100) comprising: a generator (10), which is configured to generate an initial loss function (L_I) comprising at least one initial covariance matrix (ICM) defining a model of correlated noise for each pixel of the vector-valued image; a processor (20), which is configured to provide a final loss function (L_F) comprising a set of at least one final covariance matrix (FCM) based on the initial loss function by modifying at least one submatrix and/or at least one matrix element of the initial covariance matrix (ICM); and a noise-suppressor (30), which is configured to denoise the vector-valued image using the final loss function (L_F) comprising the set of the at least one final covariance matrix (FCM).