Deep denoiser prior based deep analytic network for lensless image restoration.

Area of Science:

• Optics and Photonics
• Computer Vision
• Machine Learning

Background:

• Mask-based lensless imagers offer potential for ultra-thin devices but suffer from poor image quality due to ill-conditioned systems.
• Traditional optimization methods for lensless imaging are often complex and may not yield optimal results.

Purpose of the Study:

• To develop an end-to-end deep analytic network for improving lensless image quality.
• To address the challenge of poor visual perception in mask-based lensless imaging systems.
• To provide a universal solution for non-blind image restoration problems.

Main Methods:

• Proposed a deep analytic network that mimics traditional optimization processes.
• Integrated analytic updates with a deep denoiser for progressive image quality enhancement.
• Mathematically proved the convergence of the proposed network.
• Applied the method to general inverse problems, demonstrating its universality in non-blind restoration.

Main Results:

• The deep analytic network progressively improves lensless image quality over several iterations.
• Mathematical convergence was proven and experimentally verified.
• The method demonstrated superior performance compared to existing state-of-the-art techniques in five deblurring experiments.

Conclusions:

• The proposed deep analytic network effectively enhances lensless image quality.
• The method offers a universal and superior solution for non-blind image restoration.
• This approach holds significant promise for advancing mask-based lensless imaging applications.