Multi-Phase Deconvolution for Image Artifact Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deconvolution methods for image deblurring are complex, slow, and often produce images with strong artifacts like ringing and noise magnification, leading to unnatural textures and blurry areas, especially when dealing with spatially uniform blur and extreme brightness regions.
Innovation Solution
A multi-phase deconvolution method using different regularization masks and prior distributions, such as Laplacian, Hyper-Laplacian, and Gaussian priors, to separately reconstruct main edges and texture, along with adaptive spatial masks to handle outlier regions and color channels, resulting in high-quality deblurred images with reduced artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deconvolution methods are used to deblur images, then image sharpness is improved, but ringing artifacts and noise magnification occur
Solution Approach 1:
The patent changes the regularization parameter from L2 norm to L1 norm in the cost function. This parameter change fundamentally alters the optimization behavior to produce sparse solutions that suppress ringing artifacts and noise magnification while maintaining edge sharpness in deblurred images
Solution Approach 2:
The patent applies different regularization strengths to different regions of the image based on local characteristics. By adapting the regularization parameter locally according to image content, it maintains sharp edges in important regions while suppressing artifacts in other areas
2Object-generated harmful factors
If regularization terms are added to reduce ringing artifacts, then artifact suppression is improved, but fine texture is lost
Solution Approach 1:
The patent changes the regularization norm from L2 to L1, which fundamentally alters the solution sparsity characteristics. This enables the method to suppress ringing artifacts while preserving fine texture details that would otherwise be lost in traditional regularized least squares approaches
Solution Approach 2:
The patent employs adaptive regularization where the regularization strength dynamically adjusts based on local image characteristics. This dynamic adaptation allows the method to suppress artifacts in smooth regions while preserving fine textures in detailed regions
3Productivity
If simple deconvolution methods are used, then computational speed is improved, but image quality deteriorates
Solution Approach 1:
The patent changes the optimization objective from minimizing L2 error to minimizing L1 error with sparsity constraints. This parameter change enables the use of efficient proximal gradient descent algorithms that converge faster than traditional methods while producing higher quality results with reduced artifacts
Data Source
AI summary
A method for deblurring a blurry image (400) includes the steps of: performing a first phase of deconvolution (202) with a first phase regularization spatial mask (300) to reconstruct the main edges and generate a first phase latent sharp image (404) having reconstructed main edges; and performing a second phase of deconvolution (204) with a second phase regularization spatial mask (304) to reconstruct the texture and generate a second phase latent sharp image (406). The second phase regularization spatial mask (304) can be different from the first phase regularization spatial mask (300).