Deep Learning, Neural Networks and Holographic Processing I (DTh1D)

Presider: Xin Fan, Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences
Authentication Required

Access to this content is only available to registered attendees. Please login now to access the live content.

3:00 - 3:30
(UTC - 07:00)

Review of Deep Learning Based De-noising Algorithms for Phase Imaging and Applications to High-speed Coherent Imaging (DTh1D.1)
Presenter: Silvio Montresor, Université du Maine

We presents a review of deep-learning based algorithms dedicated to the processing of the speckle noise in phase imaging with a focus on the decorrelation phase noise. Applications to high-speed coherent imaging are discussed.

Authors:Silvio Montresor, Université du Maine / Marie Tahon, Université du Maine / Pascal Picart, Université du Maine

  Paper

3:30 - 3:45
(UTC - 07:00)

Reduced CNNs architectures applied to phase maps corrupted with speckle noise (DTh1D.2)
Presenter: Silvio Montresor, Le Mans Université

This paper addresses the problem of phase images corrupted with speckle noise. DnCNN residual networks with different depths were built and trained with various holographic noisy phase data to realistic conditions. The best results are obtained using a network with only 4 residual blocks and trained with a wide range of noisy speckle patterns.

Authors:Silvio Montresor, Le Mans Université / Marie Tahon, Le Mans Université / Pascal Picart, Le Mans Université

  Paper

3:45 - 4:00
(UTC - 07:00)

Raw holograms based machine learning for cancer cells classification in microfluidics (DTh1D.3)
Presenter: Mattia Delli Priscoli, University of Salerno

We investigate the ability of machine learning to provide an accurate classification of cancer cell in microfluidics when only raw digital holograms are used as input data. Comparison among different learning strategies is addressed.

Authors:Mattia Delli Priscoli, University of Salerno / Pasquale Memmolo, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Gioele Ciaparrone, University of Salerno / Vittorio Bianco, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Francesco Merola, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Lisa Miccio, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Francesco Bardozzo, University of Salerno / Daniele Pirone, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Martina Mugnano, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Flora Cimmino, CEINGE Biotecnologie Avanzate / Mario Capasso, CEINGE Biotecnologie Avanzate / Achille Iolascon, CEINGE Biotecnologie Avanzate / Pietro Ferraro, CNR - Istituto di Scienze Applicate e Sistemi Intelligenti "Eduardo Caianiello" / Roberto Tagliaferri, University of Salerno

  Paper

4:00 - 4:15
(UTC - 07:00)

Limited-view Cone Beam CT reconstruction using 3D Patch-based Supervised and Adversarial Learning (DTh1D.4)
Presenter: Anish Lahiri, University of Michigan

We present a novel multi-stage algorithm for CBCT reconstruction from very limited projections. Our proposed method uses 3D patch-basedsupervised and adversarial learning from scarce training data, combined with physics (forward) models and statistical priors.

Authors:Anish Lahiri, University of Michigan / Marc Klasky, Los Alamos National Laboratory / Jeffrey Fessler, University of Michigan / Saiprasad Ravishankar, Michigan State University

  Paper

4:15 - 4:30
(UTC - 07:00)

Noise influence on DeepDensity: convolutional neural network for local fringe density map estimation (DTh1D.5)
Presenter: Maria Cywinska, Warsaw University of Technology

Convolutional neural network based, fast and accurate local fringe density map estimation by DeepDensity significantly enhances full-field optical measurement techniques, e.g., holographic microscopy. The numerical capabilities of the proposed algorithmic solution in the case of the presence of noise were studied.

Authors:Maria Cywinska, Warsaw University of Technology / Filip Brzeski, Warsaw University of Technology / Wiktor Krajnik, Warsaw University of Technology / Krzysztof Patorski, Warsaw University of Technology / Maciej Trusiak, Warsaw University of Technology

  Paper

4:30 - 4:45
(UTC - 07:00)

Learning-based method for speckle noise reduction of numerically reconstructed holograms without recording experimental datasets for its training (DTh1D.6)
Presenter: Carlos Trujillo, Universidad EAFIT

A convolutional autoencoder for speckle noise reduction of numerically reconstructed holograms is presented. The neural network is trained with a generic open dataset whose images are preprocessed to emulate speckle noise. Experimental validation is provided.

Authors:Sergio Jurado, Universidad EAFIT / Carlos Trujillo, Universidad EAFIT

  Paper

4:45 - 5:00
(UTC - 07:00)

Complex field recovery from on-axis digital hologram using deep learning (DTh1D.7)
Presenter: Yeon-Gyeong Ju, Inha University

Conventional on-axis holography requires multiple holograms to obtain a complex field. We propose a deep learning network that recovers complex field preserving the depth information from a single on-axis hologram of diffusive objects.

Authors:Yeon-Gyeong Ju, Inha University / Jae-Hyeung Park, Inha University

  Paper