Efficient restoration of diffraction-induced space-variant blur in stacked microlens array scanning imaging system

Area of Science:

• Optics and Photonics
• Image Processing
• Computational Imaging

Background:

• Microlens array scanning imaging systems exhibit space-variant point spread functions due to diffraction, causing significant image blur.
• Traditional deconvolution methods struggle to effectively restore images degraded by space-variant blur.

Purpose of the Study:

• To develop an efficient and accurate space-variant image deblurring model for microlens array imaging systems.
• To introduce a deep unfolding network, MSVKNet, for rapid and effective image restoration.

Main Methods:

• Representing the image degradation process using sparse matrices within an alternating direction method of multipliers optimization framework.
• Designing and implementing the Multiple Space-Variant Kernel Network (MSVKNet), a deep unfolding network.
• Comparing MSVKNet performance against a Total Variation (TV) priors and Conjugate Gradient (CG) iteration (TV-CG) method.

Main Results:

• MSVKNet demonstrates restoration performance comparable to or exceeding the TV-CG method across various system parameters.
• MSVKNet achieves an inference speedup of nearly three orders of magnitude compared to TV-CG.
• Experimental validation on a dual microlens array system confirms the method's accuracy and practical utility.

Conclusions:

• The proposed MSVKNet effectively addresses space-variant image deblurring challenges in microlens array systems.
• MSVKNet offers a significant speed advantage for real-time or near-real-time image restoration applications.
• The deep unfolding approach provides a robust and practical solution for improving image quality in computational imaging.