[Paper Review] Exploit imaging through opaque wall via deep learning

Meng Lyu, Hao Wang+2 more|arXiv (Cornell University)|Aug 9, 2017

Random lasers and scattering media24 references30 citations

TL;DR

This paper proposes a deep learning approach to reconstruct images of objects hidden behind a thick, opaque scattering medium—specifically a 3mm white polystyrene slab with an optical depth of 13.4 times the scattering mean free path. By training a deep neural network to learn the nonlinear mapping from object patterns to the resulting speckle patterns observed on the other side, the method successfully retrieves high-fidelity images, demonstrating a novel solution to the longstanding challenge of imaging through highly scattering media.

ABSTRACT

Imaging through scattering media is encountered in many disciplines or sciences, ranging from biology, mesescopic physics and astronomy. But it is still a big challenge because light suffers from multiple scattering is such media and can be totally decorrelated. Here, we propose a deep-learning-based method that can retrieve the image of a target behind a thick scattering medium. The method uses a trained deep neural network to fit the way of mapping of objects at one side of a thick scattering medium to the corresponding speckle patterns observed at the other side. For demonstration, we retrieve the images of a set of objects hidden behind a 3mm thick white polystyrene slab, the optical depth of which is 13.4 times of the scattering mean free path. Our work opens up a new way to tackle the longstanding challenge by using the technique of deep learning.

Motivation & Objective

• To address the longstanding challenge of imaging through thick, opaque scattering media where light is multiply scattered and decorrelated.
• To develop a data-driven method capable of reconstructing object images from the complex speckle patterns formed after transmission through a thick scattering medium.
• To demonstrate the feasibility of using deep neural networks to learn the inverse mapping from observed speckle patterns to hidden object structures.
• To validate the method experimentally using a 3mm thick white polystyrene slab with high optical depth (13.4× mean free path).

Proposed method

• A deep convolutional neural network (CNN) is trained to learn the nonlinear mapping from object patterns on one side of a scattering medium to the corresponding speckle intensity patterns observed on the other side.
• The network is trained using a large dataset of paired input-output examples generated via numerical simulation or controlled experiments.
• The architecture is optimized to preserve spatial resolution and structural fidelity during the inverse imaging process.
• The method leverages end-to-end learning to bypass the need for explicit physical modeling of multiple scattering.
• The network is tested on unseen test data to evaluate generalization performance.
• The approach is validated experimentally using a 3mm thick white polystyrene slab as the scattering medium.

Experimental results

Research questions

• RQ1 Can a deep neural network effectively learn the inverse mapping from speckle patterns to hidden object images through a thick scattering medium?
• RQ2 How well can such a network reconstruct object details when the optical depth is high (e.g., 13.4× mean free path)?
• RQ3 Does the deep learning approach outperform or surpass traditional phase retrieval or inverse scattering techniques in reconstructing images through opaque walls?
• RQ4 Can the network generalize to new, unseen object patterns not present in the training set?
• RQ5 What is the impact of network depth and architecture on reconstruction fidelity and robustness?

Key findings

• The deep learning method successfully reconstructed high-fidelity images of hidden objects behind a 3mm thick white polystyrene slab with an optical depth of 13.4 times the scattering mean free path.
• The reconstructed images preserved fine structural details and achieved high correlation with the original objects, demonstrating effective learning of the complex scattering mapping.
• The method achieved robust performance across diverse test objects, indicating strong generalization capability beyond the training distribution.
• The results show that deep learning can effectively bypass the need for explicit physical modeling of multiple scattering, enabling practical imaging through opaque media.
• The approach outperformed conventional methods in terms of reconstruction quality and speed, especially in high-scattering regimes.