Image Deblurring via Enhanced Low-Rank Prior

Area of Science:

• Computer Vision
• Image Processing

Background:

• Low-rank matrix approximation is a powerful tool for various vision tasks.
• Blind image deblurring remains a challenging problem in image processing.

Purpose of the Study:

• To propose a novel low-rank prior for blind image deblurring.
• To enhance existing low-rank methods for improved deblurring performance.

Main Methods:

• A novel low-rank prior is developed by analyzing similar patches from both the blurry image and its gradient map.
• Weighted nuclear norm minimization is employed to refine the low-rank prior, focusing on dominant edges.
• The method is evaluated for both uniform and non-uniform deblurring scenarios.

Main Results:

• The proposed low-rank prior effectively reduces blur while preserving crucial edge information.
• The enhanced prior demonstrates superior performance in kernel estimation compared to simpler models.
• Experimental results show the algorithm performs favorably against state-of-the-art deblurring methods.

Conclusions:

• The enhanced low-rank prior offers a significant advancement in blind image deblurring.
• The approach provides a robust solution for both uniform and non-uniform deblurring.
• This method holds promise for various computer vision applications requiring high-quality image restoration.