Digital Image Restoration With Deconvolution Before Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital image restoration methods fail to effectively remove acquisition and compression noise, leading to visually distracting artifacts and impairing further image exploitation, particularly in satellite imagery.
Innovation Solution
A digital image restoration method involving decompression with restitution of instrumental acquisition noise, followed by variance stabilization transformation, stationary noise denoising, and deconvolution, optionally with image fusion, to improve image quality by modifying noise distribution and reducing artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If image compression is applied to reduce data size, then storage and transmission efficiency is improved, but compression artifacts and noise are introduced that degrade image quality
Solution Approach 1:
The patent applies deconvolution as a preliminary step before denoising to remove compression artifacts. By restoring the image structure first through deconvolution with the compression kernel, subsequent denoising operations can more effectively remove noise while preserving image details, thus resolving the quality degradation caused by compression.
Solution Approach 2:
The patent introduces an intermediate deconvolved image as a mediator between the compressed image and the final denoised output. This intermediate representation separates the artifact removal and noise removal processes, allowing each to be optimized independently while working together to restore overall image quality.
2Manufacturing precision
If deconvolution is applied to remove compression artifacts, then image sharpness is improved, but noise is amplified in the process
Solution Approach 1:
The patent performs deconvolution first to restore image sharpness and structure, accepting temporary noise amplification. Then, a subsequent denoising step is applied to the deconvolved image to remove the amplified noise while preserving the restored sharpness, thus achieving both goals sequentially.
Solution Approach 2:
The patent creates a continuous processing chain where deconvolution and denoising are applied in sequence without interruption. The output of deconvolution becomes the input for denoising, ensuring that the useful action of artifact removal is maintained while continuously managing noise levels through the combined process.
3Manufacturing precision
If denoising is applied to remove acquisition noise, then image quality is improved, but image details and edges may be blurred
Solution Approach 1:
The patent applies deconvolution before denoising to restore image edges and details first. By establishing the correct image structure and sharpness beforehand, the subsequent denoising process can more selectively remove noise while preserving the restored edges and fine details, reducing information loss.
Solution Approach 2:
The patent replaces simple noise filtering with a more sophisticated two-step process involving deconvolution and variance-stabilized denoising. This substitution uses mathematical modeling of the noise characteristics and image formation process to achieve better noise removal while preserving details, rather than relying on basic mechanical filtering.
4Object-generated harmful factors
If standard denoising methods are applied to compressed images, then noise is reduced, but compression artifacts remain and image sharpness is lost
Solution Approach 1:
The patent reverses the conventional order by applying deconvolution before denoising. This preliminary restoration of image structure through deconvolution ensures that when denoising is subsequently applied, the image sharpness and edges are already restored, so noise removal does not compromise image quality.
Solution Approach 2:
The patent introduces deconvolution as an intermediary process between compression and denoising. This intermediary step addresses the compression artifacts that standard denoising methods cannot handle, creating a bridge that allows subsequent denoising to work more effectively without losing image sharpness.
Data Source
Figure 1~3
Figure 4
Figure 5
AI summary
The invention concerns a method and a device for digital image restoration. The digital image to be restored comes from an initial image acquired by an image acquisition device having an associated acquisition instrumental noise. The method comprises the following steps: - obtaining (30) an intermediate digital image with restoration of the acquisition instrumental noise, - denoising (32) the intermediate digital image in order to obtain a denoised intermediate digital image, - deconvolution (34) of the denoised digital image in order to obtain a restored digital image.