Image Denoising via Adaptive Noise Variance Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image denoising methods, such as BM3D, apply a single noise variance value across the entire image, which can lead to blurring when actual noise variance varies within the image, especially in low light conditions.
Innovation Solution
The method determines assessed noise variance for each patch or patch set based on content characteristics, allowing for dynamic noise reduction by transforming the patch set into the frequency domain and applying noise reduction techniques like hard-thresholding or Wiener filtering, before inverse transforming back to the spatial domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a single noise variance value is applied across the entire image, then the denoising process is simple and fast, but the image quality deteriorates with blurring when actual noise variance varies within the image
Solution Approach 1:
The image is divided into multiple patches, and each patch is further divided into sub-regions for independent noise variance assessment. This segmentation allows the application of different noise variance values to different regions, improving denoising quality by adapting to local noise characteristics while maintaining computational feasibility through localized processing.
Solution Approach 2:
The patent implements local quality assessment by determining noise variance separately for different patches and sub-regions based on their specific content characteristics. This enables the denoising process to apply appropriate noise variance values to each local region, preventing blurring in areas with varying noise levels while maintaining overall image quality.
2Manufacturing precision
If noise variance is determined for each patch based on content characteristics, then denoising quality is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary assessment of noise variance for each patch and sub-region before the actual denoising operation. By pre-determining the noise variance values based on content characteristics, the subsequent denoising process can proceed efficiently with known parameters, reducing overall processing time despite the initial assessment step.
Solution Approach 2:
The patent applies noise variance determination to selected patches and sub-regions rather than uniformly processing the entire image. This partial action approach focuses computational resources on regions where content characteristics indicate varying noise levels, improving denoising quality where needed while reducing unnecessary processing in uniform regions.
Data Source
AI summary
A method for denoising images by block-matching three-dimensional (BM3D) method is disclosed in the present invention. Embodiments of the present invention are used to improve the quality of captured images. Instead of using the same noise variance to denoise all patches of an image, each patch is processed based on a particular assessed noise variance. The assessed noise variance of one reference patch is determined based on noise variance associated with the patch set or based on content characteristics associated with the patch set. The patch set is obtained by block-matching to find similar patches of the reference patch. Noise reduction in frequency domain is applied to the patch set according to the assessed noise variance of the reference patch. The determining of the assessed noise variance can be performed in spatial domain or in frequency domain.