Deeply coded aperture for lensless imaging.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Lensless imaging offers advantages in miniaturization and cost but faces challenges in image reconstruction.
• Coded apertures modulate light wavefronts to encode spatial information, aiding reconstruction.
• Deep learning, particularly convolutional neural networks (CNNs), has shown promise in solving complex inverse problems like image reconstruction.

Purpose of the Study:

• To present a method for jointly designing a coded aperture and a CNN for improved object reconstruction.
• To integrate the coded aperture as the first layer within a deep learning framework.
• To experimentally validate the co-optimization approach and compare its performance.

Main Methods:

• A deep learning framework was utilized to co-optimize the coded aperture and the reconstruction CNN.
• The coded aperture was implemented as the initial convolutional layer of the CNN.
• The proposed method was experimentally demonstrated using a fully convolutional network architecture.

Main Results:

• The joint design approach enabled effective object reconstruction from single-shot lensless measurements.
• Experimental validation confirmed the efficacy of the co-optimization strategy.
• Performance comparison showed advantages over traditional coded apertures, such as the modified uniformly redundant array.

Conclusions:

• Jointly designing coded apertures and CNNs offers a powerful approach for lensless imaging.
• This deep learning-integrated method significantly enhances object reconstruction capabilities.
• The proposed framework provides a new direction for advanced computational imaging systems.