Lensless cameras using a mask based on almost perfect sequence through deep learning.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Deep Learning

Background:

• Mask-based lensless imaging offers advantages in size and cost for various applications.
• However, inherent ill-posed inverse problems lead to low-resolution and poor-quality reconstructed images.
• Existing methods struggle to overcome these limitations effectively.

Purpose of the Study:

• To develop an advanced image reconstruction method for mask-based lensless imaging.
• To enhance the resolution and quality of reconstructed images using a novel approach.
• To leverage deep learning and physical imaging models for superior performance.

Main Methods:

• Utilized a mask with excellent autocorrelation properties based on an almost perfect sequence.
• Proposed a Learned Analytic solution Net (LAN) within an unrolled optimization framework.
• Integrated a physical imaging model with deep learning for image reconstruction.

Main Results:

• Achieved high-quality image reconstruction at a resolution of 512x512 pixels.
• Demonstrated excellent performance in both visual quality and objective evaluations.
• The proposed network effectively combines physical principles with deep learning insights.

Conclusions:

• The developed Learned Analytic solution Net significantly improves lensless imaging reconstruction.
• The use of an almost perfect sequence mask enhances imaging capabilities.
• This approach offers a promising solution for high-quality, low-cost imaging systems.