Joint-learning-based coded aperture compressive temporal imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Signal Processing

Background:

• Coded aperture compressive temporal imaging (CACTI) is an advanced technique for capturing high-speed dynamic scenes.
• Reconstructing 3D dynamic scenes from 2D compressive measurements presents significant challenges in sampling and inverse problems.

Purpose of the Study:

• To develop a flexible, efficient, and high-quality reconstruction method for CACTI.
• To jointly optimize the mask, relaxation, and regularization strategies for improved video reconstruction.

Main Methods:

• A novel Landweber iterative reconstruction method is proposed, integrating deep learning for parameter selection.
• Multiple denoisers are employed to enhance sparse representation and regularization during iterative reconstruction.
• The method jointly learns optimal binary mask, relaxation, and regularization strategies.

Main Results:

• The proposed method demonstrates superior performance in reconstructing high-speed dynamic scenes.
• Experimental results validate the effectiveness of the joint learning approach for mask and parameter optimization.
• The integration of deep learning successfully addresses the challenging parameter selection in the Landweber method.

Conclusions:

• The developed Landweber iterative method offers a flexible, efficient, and high-quality solution for CACTI video reconstruction.
• Jointly learning optimization strategies and incorporating deep learning significantly improves reconstruction accuracy.
• This approach advances the capabilities of compressive sensing techniques for dynamic scene imaging.