Extended Depth-Of-Field In Holographic Image Reconstruction Using Deep Learning-Based Auto-Focusing And Phase-Recovery

Summary

UCLA researchers in the Department of Electrical Engineering have developed a novel deep learning-based algorithm that digitally reconstructs images from holography over an extended depth of field.

Background

Holographic imaging has many applications in the fields of engineering, research and medicine. A holography encodes the 3D information of a sample. However, it is time-consuming and cumbersome to digitally decode the original sample image from its hologram. This process requires auto-focusing and phase recovery, which are complex, computationally heavy and specific to the imaging set-up. This leads to limitations in the depth-of-field (DOF) in image reconstruction, which in turn limits the application of this imaging modality.

Innovation

A novel convolutional neural network (CNN)-based approach was developed to digitally decode holograms. It simultaneously performs auto-focusing and phase recovery to significantly extend the DOF of holographic image reconstruction. This CNN was trained to quickly reconstruct an in-focus image of a sample over an extended DOF from a single input of back-propagated hologram of a 3D sample. It improves upon the algorithm time complexity of existing methods and is non-iterative. It can also be applied to other imaging modalities to extend their DOF.

Applications

• Digital holography 
• Other imaging modalities such as florescence imaging

Advantages

• Fast 
• Non-iterative 
• Extended DOF 
• Widely applicable

Patent Status