Blind multi-Poissonian image deconvolution with sparse log-step gradient prior.

Area of Science:

• Image processing
• Computational imaging
• Scientific visualization

Background:

• Blind image deconvolution is crucial for astronomical and microscopy imaging.
• Poisson noise in images presents significant challenges for deconvolution, especially at high noise levels.
• Restoring single blurred images with high noise is often ill-posed and yields unsatisfactory results.

Purpose of the Study:

• To develop a robust method for high-quality blind multi-image deconvolution.
• To address challenges posed by Poisson noise in image restoration.
• To improve image quality in applications like astronomy and fluorescence microscopy.

Main Methods:

• Designed a novel sparse log-step gradient prior incorporating logarithm and step functions for image gradient regularization.
• Formulated the blind multi-image deconvolution problem by combining the prior with the Poisson distribution.
• Employed variable splitting and Lagrange multiplier methods to solve the problem by converting it into solvable sub-problems.
• Developed a non-blind multi-image deconvolution algorithm based on the log-step gradient prior for final image restoration.

Main Results:

• The proposed sparse log-step gradient prior effectively suppresses artifacts arising from ill-posedness.
• The developed algorithm achieves high-quality restored images from multi-blurred inputs.
• Experimental results on synthetic and real-world data demonstrate competitive performance against state-of-the-art methods.

Conclusions:

• The proposed method offers a powerful solution for blind multi-image deconvolution in the presence of Poisson noise.
• The log-step gradient prior significantly enhances image restoration quality and artifact suppression.
• This technique holds promise for improving image quality in demanding scientific imaging applications.