Residual-Image Denoising for Edge-Preserving Image Sharpening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing denoising methods for image sensors result in blurry images with poor definition, insufficient denoising strength at strong edges, and edge-preserving properties, leading to aliasing issues.
Innovation Solution
A denoising method that involves filtering residual images at different resolutions, enhancing weak edges and attenuating strong edges, followed by reconstructing the image using a guide image and a noisy image to enhance local contrast and details, while reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If spatial domain denoising is performed on images, then signal-to-noise ratio is improved, but image clarity and definition deteriorate
Solution Approach 1:
The patent segments the image into multiple residual images at different resolutions (first residual image, second residual image, etc.). Each residual image is processed independently through filtering and sharpening operations, allowing selective enhancement of different frequency components without blurring the entire image. This segmentation enables the denoising algorithm to work on localized regions while preserving global image quality.
Solution Approach 2:
The patent applies different processing strengths to different regions of the image based on edge detection. Strong edge regions receive attenuation to reduce noise, while weak edge regions receive enhancement to improve definition. This local quality approach allows the algorithm to adaptively adjust processing parameters in different areas, maintaining image clarity while reducing noise where needed.
2Stability of the object's composition
If edge-preserving denoising algorithm is used, then edge structure is maintained, but denoising strength at strong edges becomes insufficient
Solution Approach 1:
Instead of applying strong denoising to the entire image and then preserving edges, the patent inverts the approach by first identifying strong edge regions and applying attenuation specifically to those regions. The residual images are processed to enhance edges and details, then the results are combined. This inversion allows the algorithm to maintain edge structure while achieving strong denoising effect at critical edge locations.
Solution Approach 2:
The patent dynamically changes processing parameters based on edge strength detection. Regions with strong edges receive different processing parameters (attenuation) compared to regions with weak edges (enhancement). This parameter change approach enables the algorithm to adaptively adjust denoising strength at strong edges while maintaining overall edge structure, resolving the contradiction between edge preservation and denoising effectiveness.
3Reliability
If filtering is applied to reduce noise, then noise is reduced, but image details and edges become blurred
Solution Approach 1:
The patent segments the image into multiple residual images at different resolutions, allowing selective filtering at each level. The first residual image (difference between original and low-pass filtered image) contains high-frequency details, while the second residual image contains mid-frequency information. By processing each residual image separately and combining the results, the algorithm can reduce noise in low-frequency regions while preserving high-frequency details and edges.
Solution Approach 2:
The patent transforms the image processing from a single-plane operation to a multi-dimensional approach by creating residual images at different resolutions. This dimensional decomposition allows the algorithm to apply filtering operations in the frequency domain while preserving spatial details. The combination of residual images at different levels reconstructs the denoised image with enhanced details and reduced noise.
Data Source
AI summary
A denoising method, a denoising apparatus, an electronic device and a computer-readable storage medium are provided. One aspect includes performing filtering on a residual image of an image to generate a filtered image. The residual image may be a difference value of the image at different resolutions. The image may correspond to at least two residual images, and the at least two residual images have a preset layer sequence relationship therebetween. An enhancement on a weak edge region in the filtered image, and an attenuation on a strong edge region in the filtered image may be performed, so as to generate a sharpened image. A target image according to a reconstructed image and the sharpened image may be obtained. The reconstructed image may be a superposition of all residual images whose layer sequence is before the residual image. Sharpening may be performed during the superposition of reconstruction.