Blind Image Deblurring Using a Non-Linear Channel Prior Based on Dark and Bright Channels

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Blind image deblurring aims to restore images without prior knowledge of the blur kernel.
• Existing dark and extreme channel priors are effective but fail when images lack extreme pixel values.
• This limitation hinders blur kernel estimation in specific deblurring scenarios.

Purpose of the Study:

• To propose a novel and robust Non-Linear Channel (NLC) prior for improved blur kernel estimation.
• To address the limitations of existing priors in blind image deblurring.
• To enhance the accuracy of image restoration in challenging deblurring cases.

Main Methods:

• Introduced a Non-Linear Channel (NLC) prior based on the principle that blurring increases the dark-to-bright channel ratio.
• Developed an efficient Projected Alternating Minimization (PAM) algorithm.
• Integrated approximate strategy, half-quadratic splitting, and FISTA within the PAM algorithm for optimization.

Main Results:

• The NLC prior effectively aids blur kernel estimation where other methods fail.
• The PAM algorithm efficiently handles the complex optimization model introduced by the NLC prior.
• State-of-the-art deblurring performance was achieved on synthetic and real-world blurry images.

Conclusions:

• The proposed NLC prior offers a robust solution for blind image deblurring, particularly in challenging scenarios.
• The developed PAM algorithm provides an efficient means to implement the NLC prior.
• The method demonstrates significant improvements in image restoration quality and kernel estimation accuracy.