Blur Removal Via Blurred-Noisy Image Pair.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Real-world images often suffer from complex, mixed blur (space-variant and space-invariant) that is difficult to model mathematically.
• Existing deblurring methods frequently require accurate estimation of blur kernels, which is challenging for complex blurs.

Purpose of the Study:

• To propose a novel image deblurring method that bypasses the need for blur kernel estimation.
• To leverage a pair of easily acquired low-light images for effective deblurring.

Main Methods:

• Utilized a blurred image (low shutter speed, low ISO) and a noisy image (high shutter speed, high ISO).
• Extended the Gaussian mixture model (GMM) to model patch intensity distributions using optical flow for correspondence.
• Employed the Expectation Maximization (EM) algorithm for GMM parameter estimation and incorporated a bilateral term for feature preservation.
• Added a detail layer for final image refinement.

Main Results:

• The proposed method successfully deblurs images with complex blur types.
• Achieved superior performance compared to state-of-the-art techniques in terms of robustness and visual quality.
• Demonstrated effectiveness on both synthetic and real-world datasets.

Conclusions:

• The novel approach offers a robust and effective solution for image deblurring without explicit blur kernel estimation.
• The method shows significant potential for practical applications, especially in low-light imaging scenarios.