Speaker: Dana Berman

Ph.D. student under the supervision of Prof. Shai Avidan and Prof. Tali Treibitz

Wednesday, January 10^th, 2018 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Haze Lines for single image dehazing and color restoration

Abstract

Images taken in scattering media, such as haze, fog, and underwater, often look faded and lack contrast. The degradation is different for every pixel and depends on the distance of the scene point from the camera.

For terrestrial images, this dependency is expressed in the transmission coefficients, which control the scene attenuation and amount of haze in every pixel. Previous methods solve the single image dehazing problem using various patch-based priors. We, on the other hand, propose an algorithm based on a new and non-local prior. The algorithm relies on the assumption that colors of a haze-free image are well approximated by a few hundred distinct colors, which form tight clusters in RGB space. Our key observation is that pixels in a given cluster are often non-local, i.e., they are spread over the entire image plane and are located at different distances from the camera. In the presence of haze these varying distances translate to different transmission coefficients. Therefore, each color cluster in the clear image becomes a line in RGB space, that we term a haze-line. Using these haze-lines, our algorithm recovers both the distance map and the haze-free image.

We show how to expand the model to restore the colors of underwater images, by incorporating spectral dependency of the attenuation coefficients.

The algorithm is linear in the size of the image, deterministic and requires no training. It performs well on a wide variety of images and is competitive with other state-of-the-art methods.

You are invited to attend a lecture

All-optical Injection and Control of Currents in Carbon Films

By:

Yuri Svirko

Professor, Department of Physics and Mathematics,

University of Eastern Finland,

Joensuu, Finland

Abstract

Strong and broadband light absorption in graphene and other ultrathin carbon materials allows one to achieve high carrier densities essential for observation of nonlinear optical phenomena and makes graphene a unique playground for studying many-body effects. Being of strong fundamental importance, these effects also open a wide range of opportunities in photonics and optoelectronics. It is possible in particular to make use of strong photon-drag effect to generate and optically manipulate ensembles of charge carriers via direct transfer of momentum from photons of incident laser beam to excited electrons in unbiased sample.

Direction and amplitude of the photon drag currents in graphene and other carbon films are determined by polarization, incidence angle and intensity of the obliquely incident laser beam. In the two-beam experiment, when sample was irradiated with two mirror-reflected beams, one can either suppress or enhance current produced by each of the beams depending on the time delay between excitation laser pulses. Since the drag current strongly depends on the polarization of the excitation beam, the net current can also be tuned by rotating the polarization plane azimuth of the first beam. The observed phenomenon is solely based on the second order nonlinear optical effect and provides full and non-contact control of the direction, amplitude as well as temporal profile of the injected ultrafast photocurrent. This opens a very interesting opportunity to generate THz pulses with predicted waveform and gives additional insight on dynamics of hot carriers in graphene and other ultrathin carbon films.

On Thursday, December 21, 2017, 15:00

Room 011, Kitot building

Speaker: Neta Gertzovsky

M.Sc. student under the supervision of Prof. Nahum Kiryati and Dr. Arnaldo Mayer

Wednesday, December 6^th 2017 at 15:00

Room 011, Kitot Bldg., Faculty of Engineering

Kidney Segmentation and Renal Lesion Detection in 3D CT

Abstract

Renal cysts are common in aging kidneys, and are usually found incidentally in patients undergoing abdominal imaging for other reasons. Although most cysts are benign, they require expert examination as some of them may either indicate the presence of a malignancy or evolve into one. So far, the proposed algorithms for the analysis and detection of renal cysts have been either semi-automatic or evaluated on fairly small data-sets. Here we present a fully automatic method to segment kidneys and to detect simple renal cysts. A fully convolutional neural network (FCN) is employed for segmentation of the kidneys. A combined 3D distance map of the kidneys and surrounding fluids provides initial candidates for cysts. Then, a convolutional neural network (CNN) classifies the candidates as cysts or non-cyst objects. Performance was evaluated on 52 randomly selected volumetric CT scans with 70 cysts annotated by an experienced radiologist, with promising results.

Another type of renal lesions are cancerous renal tumors. Though also usually an incidental finding, they are malignant and often fatal. Early detection of such tumors is highly advantageous for recovery. Using the same methods and similar data we attempted to develop a fully automatic system for detection of cancerous renal tumors. We present our experiments and preliminary results and discuss the steps we deem required to undertake this challenge.