Lensless computational imaging with a hybrid framework of holographic propagation and deep learning.

Abstract

Lensless imaging has attracted attention as it avoids the bulky optical lens. Lensless holographic imaging is a type of a lensless imaging technique. Recently, deep learning has also shown tremendous potential in lensless holographic imaging. A labeled complex field including real and imaginary components of the samples is usually used as a training dataset. However, obtaining such a holographic dataset is challenging. In this Letter, we propose a lensless computational imaging technique with a hybrid framework of holographic propagation and deep learning. The proposed framework takes recorded holograms as input instead of complex fields, and compares the input and regenerated holograms. Compared to previous supervised learning schemes with a labeled complex field, our method does not require this supervision. Furthermore, we use the generative adversarial network to constrain the proposed framework and tackle the trivial solution. We demonstrate high-quality reconstruction with the proposed framework compared to previous deep learning methods.