You are invited to attend a department seminar on

Advanced Techniques for Color Light Field

Reconstruction and Depth Estimation from Compressed Measurements

:By

Ofir Nabati

MSc student under the supervision of Prof. David Mendlovic and Dr.Raja Giryes

Abstract

In the last decade, the usage of digital cameras has grown exponentially for various reasons such as photography, communication and security. While conventional cameras allow for capturing the spatial information of the scene, light field

photography allows to capture the angular information as well. By doing so, it

enables for applications such as refocusing and depth estimation.

The notion of light field photography was present almost a century ago. Since

then, there has been no major breakthrough in the transition from theory into wide

usage, mainly because suggested solutions suffered heavily in terms of loss of resolution, computational time and light efficiency. One of its drawbacks is the need

for multi-lens in the imaging. To compensate that, compressed light field photography has been proposed to tackle the trade-offs between the spatial and angular

resolutions. It obtains by only one lens, a compressed version of the regular multilens system. The acquisition system consists of a dedicated hardware followed by

a decompression algorithm, which relies on the theory of compressed sensing and

sparse coding techniques. The reconstruction process usually suffers from high

computational time. In this thesis, we review various methods for reconstruction of compressed light fields and also propose a computationally efficient deep

learning based algorithm that recovers a high-quality color light field from a single coded image. Unlike previous works, we compress the color channels as well,

removing the need for a CFA in the imaging system.

Our approach outperforms existing solutions in terms of recovery quality and

computational complexity. We propose also a neural network for depth map extraction based on the decompressed light field, which is trained in an unsupervised

manner without the ground truth depth map. We also show the implementation

and performance of our algorithm in a real compressed light field camera prototype, which is significantly smaller and cheaper compared to existing commercial

light field cameras.

On Thursday, January 10, 2018, 14:00

Room 011, EE-Class Building

Speaker:  Hillel Sreter

M.Sc. student under the supervision of Dr. Raja Giryes

Sunday, December 30^th, 2018 at 15:00

Room 011, Kitot Bldg., Faculty of Engineering

Learned Convolutional Sparse Coding
 

Abstract

    We propose a convolutional recurrent sparse auto-encoder model. The model consists of a sparse encoder, which is a convolutional extension of the learned ISTA (LISTA) method, and a linear convolutional decoder.

    Our strategy offers a simple method for learning a task-driven sparse convolutional dictionary (CD) and producing an approximate convolutional sparse code (CSC) over the learned dictionary.

    We trained the model to minimize reconstruction loss via gradient decent

with back-propagation and have achieved competitive results to KSVD image denoising and to leading CSC methods in image inpainting requiring only a small fraction of their run-time.

You are invited to attend a department seminar on

Multi-lobe superoscillations and Structured illumination microscopy

:By

Niv Shapira

MSc student under the supervision of Prof. Ady Arie

Abstract

Superoscillation is a phenomenon of growing interest, describing a wave interference between frequency components of a band-limited function, that is locally oscillating faster than its highest Fourier component. In this work we study and implement methods to generate multi-lobe optical superoscillating beams, and their potential application for super-resolution structured illumination microscopy. Whereas many previous works concentrated on generating a single super-oscillating lobe, here we study methods to generate a multi-lobe function with nearly constant intensity and constant local frequency. Generating such periodic superoscillations is a challenging task, but may open the door for interesting applications which require its periodicity. We generated superoscillation patterns having 3 to 11 sub-wavelength lobes, with frequency of 20 to 40 percent above the system cut-off frequency. These patterns may be useful for structured illumination microscopy, enabling more than two-fold improvement in resolution with respect to the classical diffraction limit.

On Thursday, December 27, 2018, 15:00

Room 011, EE-Class Building