Non-Local Image Demosaicing with Noise-Stabilized Patch Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing demosaicing techniques for images captured by multi-spectral sensor arrays, such as those organized according to a Bayer pattern, often result in limited image quality, which is insufficient for applications like Earth observation in space.
Innovation Solution
A demosaicing method that includes a noise variance stabilization transformation, followed by the determination of similar patches, calculation of covariance matrices, and block-based deconstruction to estimate interpolated spectral bands, ultimately leading to improved image reconstruction quality and noise robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional demosaicing techniques are used, then the processing is simple and fast, but the image quality is limited and insufficient for space applications
Solution Approach 1:
The patent divides the image into multiple patches and processes them independently. Each patch is transformed, processed for similar patch identification, and then reassembled. This segmentation allows complex processing to be performed on manageable units without overwhelming computational complexity, improving image quality while maintaining feasibility.
Solution Approach 2:
The patent applies a noise variance stabilization transformation to the entire image before performing patch-based processing. This preliminary action prepares the data by stabilizing noise characteristics, which simplifies subsequent processing steps and improves the accuracy of similar patch identification, leading to better image quality without proportionally increasing complexity.
2Reliability
If noise variance stabilization transformation is applied, then the noise robustness is improved, but the processing time increases
Solution Approach 1:
The noise variance stabilization transformation is performed once as a preliminary step on the entire image before patch processing begins. This single transformation stabilizes noise characteristics for all subsequent operations, improving noise robustness without requiring repeated application. The one-time cost is offset by the benefits gained throughout the entire processing pipeline.
Solution Approach 2:
The transformation changes the statistical parameters of the noise distribution, converting it to a form with constant variance. This parameter change enables more effective similar patch identification and covariance matrix calculation, improving reliability while the computational overhead is managed through efficient implementation of the transformation algorithm.
3Measurement precision
If block-based deconstruction of covariance matrix is used, then the estimation accuracy of interpolated spectral bands is improved, but the computational complexity increases
Solution Approach 1:
The covariance matrix is divided into blocks corresponding to different spectral bands (red, green, blue). This segmentation allows the complex matrix operations to be performed on smaller, more manageable sub-matrices. Each block can be processed independently, reducing the computational burden while maintaining the accuracy benefits of using the full covariance structure for interpolation estimation.
Data Source
AI summary
A demosaicing method is disclosed as being applied to an initial image in order to obtain a demosaiced image, enabling an improved demosaicing due to the fact that it includes a noise variance stabilization transformation, for each reference patch: a determination of a set of similar patches, a determination of a covariance matrix for said set of similar patches, and for each similar patch, a calculation of estimates of the interpolated spectral bands according to a block-based deconstruction of the covariance matrix as a function of the phasing, a determination of the interpolated spectral bands by an aggregation of estimates of the interpolated spectral bands, and an inverse transformation of the stabilization transformation.