Blind Remote Sensing Image Deblurring Based on Overlapped Patches' Non-Linear Prior

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Remote sensing images often suffer from blur due to complex environmental factors.
• Image restoration models typically rely solely on observed blurry images without prior knowledge.
• Existing methods for image prior extraction can be computationally intensive or less effective.

Purpose of the Study:

• To develop a novel prior, the Overlapped Patches Non-Linear (OPNL) prior, for enhancing remote sensing image restoration.
• To address the limitations of existing methods in handling complex blurring in remote sensing imagery.
• To improve the accuracy and effectiveness of deblurring algorithms for remote sensing applications.

Main Methods:

• Extraction of features using partially overlapping image patches.
• Design of the OPNL prior based on the ratio of extreme pixels affected by blurring within patches.
• Development of a solving algorithm integrating projected alternating minimization (PAM), half-quadratic splitting, FISTA, and FFT.

Main Results:

• The OPNL prior demonstrates a preference for clear image characteristics during restoration.
• The developed algorithm shows excellent stability and effectiveness in experimental evaluations.
• The proposed method achieves competitive results in restoring degraded remote sensing images.

Conclusions:

• The OPNL prior is a viable and effective approach for remote sensing image restoration.
• The integrated algorithm provides a robust solution for complex image deblurring tasks.
• This research contributes to advancing the quality of remote sensing image analysis.