arXiv subjects

Ori Katz

Publications and source records attributed to Ori Katz.

At least 19 recordsLinked to original sources

arXiv2026

Leveraging target dynamics for imaging in complex media

Optical imaging in complex samples such as biological tissues is fundamentally challenging due to random light scattering that degrades resolution and contrast. When imaging realistic targets that contain natural dynamics such as flowing blood, the temporal variability introduces an additional obstacle, as the leading computational scattering-compensation methods require the target to remain stationary during a multi-frame acquisition process. Here we show that instead of struggling to perform rapid acquisitions, the target dynamics themselves can serve as an intrinsic information source for scattering compensation, replacing multiple controlled illuminations. The mathematical equivalence between target dynamics and conventional varying illumination patterns allows to demonstrate this approach in coherent holographic imaging and incoherent fluorescence microscopy using established matrix and model-based scattering-compensation techniques. Our general framework enables reconstruction of dynamic scenes with a number of acquisitions equal to the number of reconstructed frames, without the use of any spatial light modulators or illumination control.

physics.optics↗

arXiv2026

Leveraging natural fluctuations for matrix-based aberration correction in photoacoustic imaging

Photoacoustic imaging is the leading technique for deep tissue optical imaging, allowing single-shot imaging at depths. However, its resolution may be limited by acoustic aberrations, caused by natural unknown heterogeneities in the tissue speed of sound. In recent years, reflection-matrix based scattering-compensation techniques have been successfully employed in ultrasound, optics, and seismology, to computationally correct such distortions. However, they have not been adapted to photoacoustic imaging since they rely on multiple acquisitions under different controlled excitations, such as input plane-wave illuminations, which do not result in signal changes in photoacoustics. Here, we introduce a framework that enables the direct application of the state-of-the-art reflection-matrix based aberration correction techniques to photoacoustic imaging of dynamic targets. Specifically, we show that a covariance matrix analysis of a conventional set of photoacoustic frames of dynamic targets, such as flowing red blood cells in blood vessels, yields a virtual reflection-matrix that is mathematically analogous to a pulse-echo reflection-matrix, and lends itself to direct processing by conventional reflection-matrix based scattering-compensation algorithms. We validate and demonstrate the approach for photoacoustic aberration correction of vessel-mimicking targets containing flowing absorbers in both simulations and experiments.

physics.optics↗

arXiv2025

Passive acoustic non-line-of-sight localization without a relay surface

The detection and localization of a source hidden outside the Line-of-Sight (LOS) traditionally rely on the acquisition of indirect signals, such as those reflected from visible relay surfaces such as floors or walls. These reflected signals are then utilized to reconstruct the obscured scene. In this study, we present an approach that utilize signals diffracted from an edge of an obstacle to achieve three-dimensional (3D) localization of an acoustic point source situated outside the LOS. We address two scenarios - a doorway and a convex corner - and propose a localization method for each of them. For the first scenario, we utilize the two edges of the door as virtual detector arrays. For the second scenario, we exploit the spectral signature of a knife-edge diffraction, inspired by the human perception of sound location by the head-related transfer function (HRTF). In both methods, knife-edge diffraction is utilized to extend the capabilities of non-line-of-sight (NLOS) acoustic sensing, enabling localization in environments where conventional relay-surface based approaches may be limited.

cs.SD↗

arXiv2024

Beyond memory-effect matrix-based imaging in scattering media by acousto-optic gating

Imaging inside scattering media at optical resolution is a longstanding challenge affecting multiple fields, from bio-medicine to astronomy. In recent years, several groundbreaking techniques for imaging inside scattering media, in particular scattering-matrix based approaches, have shown great promise, but usually suffer from a restricted field of view (FOV) due to their reliance on the optical 'memory-effect'. Here, we demonstrate that by combining acousto-optic spatial-gating with state-of-the-art matrix-based imaging techniques, diffraction-limited imaging beyond the optical memory-effect can be achieved in a robust fashion. Specifically, we show that this can be achieved by computational processing of scattered light fields captured under scanned acousto-optic modulation. The approach can be directly utilized whenever the ultrasound focus size is of the order of the memory-effect range, independently of the scattering angle.

physics.optics↗

arXiv2024

Characterization and Exploitation of the Rotational Memory Effect in Multimode Fibers

In an ideal perfectly straight multimode fiber with a circular-core, the symmetry ensures that rotating the input wavefront leads to a corresponding rotation of the output wavefront. This invariant property, known as the rotational memory effect (RME), remains independent of the typically unknown output profile. The RME thus offers significant potential for imaging and telecommunication applications. However, in real-life fibers, this effect is degraded by intrinsic imperfections and external perturbations, and is challenging to observe because of its acute sensitivity to misalignments and aberrations in the optical setup. Thanks to processing involving a spatial light modulator, we efficiently overcome these measurement biases, allowing for precise quantification of the RME. We establish an experimental and theoretical framework for studying and manipulating the RME in multimode fibers. Theoretical predictions are consistent with experimental data and simulations, connecting the shape of the angular-dependent correlation of the RME to the geometrical properties of the core deformation. This work opens the road for accurate characterization of the distributed disorder originating from the fabrication process and calibration-less imaging in multimode fibers.

physics.optics↗

arXiv2024

Coherent Perfect Absorption of Arbitrary Wavefronts at an Exceptional Point

A Coherent Perfect Absorber (CPA) exploits the interferometric nature of light to deposit all of a light field's incident energy into an otherwise weakly absorbing sample. The downside of this concept is that the necessary destructive interference in CPAs gets easily destroyed both by spectrally or spatially detuning the incoming light field. Each of these two limitations has recently been overcome by insights from exceptional-point physics and by using a degenerate cavity, respectively. Here, we show how these two concepts can be combined into a new type of cavity design, which allows broadband exceptional-point absorption of arbitrary wavefronts. We present two possible implementations of such a Massively Degenerate Exceptional-Point absorber and compare analytical results with numerical simulations.

physics.optics↗

arXiv2024

Ptychographic lensless coherent endomicroscopy through a flexible fiber bundle

Conventional fiber-bundle-based endoscopes allow minimally invasive imaging through flexible multi-core fiber (MCF) bundles by placing a miniature lens at the distal tip and using each core as an imaging pixel. In recent years, lensless imaging through MCFs was made possible by correcting the core-to-core phase distortions pre-measured in a calibration procedure. However, temporally varying wavefront distortions, for instance, due to dynamic fiber bending, pose a challenge for such approaches. Here, we demonstrate a coherent lensless imaging technique based on intensity-only measurements insensitive to core-to-core phase distortions. We leverage a ptychographic reconstruction algorithm to retrieve the phase and amplitude profiles of reflective objects placed at a distance from the fiber tip, using as input a set of diffracted intensity patterns reflected from the object when the illumination is scanned over the MCF cores. Our approach thus utilizes an acquisition process equivalent to confocal microendoscopy, only replacing the single detector with a camera.

physics.optics↗

arXiv2023

Noninvasive megapixel fluorescence microscopy through scattering layers by a virtual reflection-matrix

Optical-resolution fluorescence imaging through and within complex samples presents a significant challenge due to random light scattering, with substantial implications across multiple fields. While significant advancements in coherent imaging through severe multiple scattering have been recently introduced by reflection-matrix processing, approaches that tackle scattering in incoherent fluorescence imaging have been limited to sparse targets, require high-resolution control of the illumination or detection wavefronts, or a very large number of measurements. Here, we present an approach that allows direct application of well-established reflection-matrix techniques to scattering compensation in incoherent fluorescence imaging. We experimentally demonstrate that a small number of conventional widefield fluorescence-microscope images acquired under unknown random illuminations can effectively construct a fluorescence-based virtual reflection matrix. This matrix, when processed by conventional matrix-based scattering compensation algorithms, allows reconstructing megapixel-scale fluorescence images, without requiring the use of spatial-light modulators (SLMs) or computationally-intensive processing.

physics.optics↗

arXiv2023

Efficient Discovery and Effective Evaluation of Visual Perceptual Similarity: A Benchmark and Beyond

Visual similarities discovery (VSD) is an important task with broad e-commerce applications. Given an image of a certain object, the goal of VSD is to retrieve images of different objects with high perceptual visual similarity. Although being a highly addressed problem, the evaluation of proposed methods for VSD is often based on a proxy of an identification-retrieval task, evaluating the ability of a model to retrieve different images of the same object. We posit that evaluating VSD methods based on identification tasks is limited, and faithful evaluation must rely on expert annotations. In this paper, we introduce the first large-scale fashion visual similarity benchmark dataset, consisting of more than 110K expert-annotated image pairs. Besides this major contribution, we share insight from the challenges we faced while curating this dataset. Based on these insights, we propose a novel and efficient labeling procedure that can be applied to any dataset. Our analysis examines its limitations and inductive biases, and based on these findings, we propose metrics to mitigate those limitations. Though our primary focus lies on visual similarity, the methodologies we present have broader applications for discovering and evaluating perceptual similarity across various domains.

cs.CV↗

arXiv2023

Representation Learning via Variational Bayesian Networks

We present Variational Bayesian Network (VBN) - a novel Bayesian entity representation learning model that utilizes hierarchical and relational side information and is particularly useful for modeling entities in the ``long-tail'', where the data is scarce. VBN provides better modeling for long-tail entities via two complementary mechanisms: First, VBN employs informative hierarchical priors that enable information propagation between entities sharing common ancestors. Additionally, VBN models explicit relations between entities that enforce complementary structure and consistency, guiding the learned representations towards a more meaningful arrangement in space. Second, VBN represents entities by densities (rather than vectors), hence modeling uncertainty that plays a complementary role in coping with data scarcity. Finally, we propose a scalable Variational Bayes optimization algorithm that enables fast approximate Bayesian inference. We evaluate the effectiveness of VBN on linguistic, recommendations, and medical inference tasks. Our findings show that VBN outperforms other existing methods across multiple datasets, and especially in the long-tail.

cs.LG↗

arXiv2023

Image scanning lensless fiber-bundle endomicroscopy

Fiber-based confocal endomicroscopy has shown great promise for minimally-invasive deep-tissue imaging. Despite its advantages, confocal fiber-bundle endoscopy inherently suffers from undersampling due to the spacing between fiber cores, and low collection efficiency when the target is not in proximity to the distal fiber facet. Here, we demonstrate an adaptation of image-scanning microscopy (ISM) to lensless fiber bundle endoscopy, doubling the spatial sampling frequency and significantly improving collection efficiency. Our approach only requires replacing the confocal detector with a camera. It improves the spatial resolution for targets placed at a distance from the fiber tip, and addresses the fundamental challenge of aliasing/pixelization artifacts.

physics.optics↗

arXiv2023

Image-guided Computational Holographic Wavefront Shaping

Optical imaging through scattering media is an important challenge in a variety of fields ranging from microscopy to autonomous vehicles. While advanced wavefront shaping techniques have offered significant breakthroughs in the past decade, current techniques still require a known guide-star and a high-resolution spatial-light-modulator (SLM), or a very large number of measurements, and are limited in their correction field-of-view. Here, we introduce a guide-star free noninvasive approach that is able to correct more than $3\cdot 10^5$ scattered modes using just $100$ holographically measured scattered random light fields. This is achieved by computationally emulating an image-guided wavefront-shaping experiment, where several 'virtual SLMs' are simultaneously optimized to maximize the reconstructed image quality. Our method shifts the burden from the physical hardware to a digital, naturally-parallelizable computation, leveraging state-of-the-art automatic-differentiation optimization tools used for the training of neural-networks. We demonstrate the flexibility and generality of this framework by applying it to imaging through various complex samples and imaging modalities, including anisoplanatic multi-conjugate correction of highly scattering layers, lensless-endoscopy in multicore fibers, and acousto-optic tomography. The versatility, effectiveness, and generality of the presented approach have great potential for rapid noninvasive imaging in diverse applications.

physics.optics↗

arXiv2023

Discovery of Single Independent Latent Variable

Latent variable discovery is a central problem in data analysis with a broad range of applications in applied science. In this work, we consider data given as an invertible mixture of two statistically independent components and assume that one of the components is observed while the other is hidden. Our goal is to recover the hidden component. For this purpose, we propose an autoencoder equipped with a discriminator. Unlike the standard nonlinear ICA problem, which was shown to be non-identifiable, in the special case of ICA we consider here, we show that our approach can recover the component of interest up to entropy-preserving transformation. We demonstrate the performance of the proposed approach in several tasks, including image synthesis, voice cloning, and fetal ECG extraction.

stat.ML↗

arXiv2023

K-space interpretation of image-scanning-microscopy

In recent years, image-scanning microscopy (ISM, also termed pixel-reassignment microscopy) has emerged as a technique that improves the resolution and signal-to-noise compared to confocal and widefield microscopy by employing a detector array at the image plane of a confocal laser scanning microscope. Here, we present a k-space analysis of coherent ISM, showing that ISM is equivalent to spotlight synthetic-aperture radar (SAR) and analogous to oblique-illumination microscopy. This insight indicates that ISM can be performed with a single detector placed in the k-space of the sample, which we numerically demonstrate.

physics.optics↗

arXiv2022

Non-Line-of-Sight Passive Acoustic Localization Around Corners

Non-line-of-sight (NLoS) imaging is an important challenge in many fields ranging from autonomous vehicles and smart cities to defense applications. Several recent works in optics and acoustics tackle the challenge of imaging targets hidden from view (e.g. placed around a corner) by measuring time-of-flight (ToF) information using active SONAR/LiDAR techniques, effectively mapping the Green functions (impulse responses) from several sources to an array of detectors. Here, leveraging passive correlations-based imaging techniques, we study the possibility of acoustic NLoS target localization around a corner without the use of controlled active sources. We demonstrate localization and tracking of a human subject hidden around the corner in a reverberating room, using Green functions retrieved from correlations of broadband noise in multiple detectors. Our results demonstrate that the controlled active sources can be replaced by passive detectors as long as a sufficiently broadband noise is present in the scene.

physics.app-ph↗

arXiv2022

Optical protection of alkali-metal atoms from spin relaxation

We present an optical technique for suppressing relaxation in alkali-metal spins using a single off-resonant laser beam. The method harnesses a physical mechanism that synchronizes Larmor precession in the two hyperfine manifolds, protecting magnetic coherence from relaxation caused by spin-exchange and other hyperfine-changing collisions. We experimentally demonstrate up to a ninefold reduction in decoherence of warm cesium vapor, achieving simultaneous protection from both spin-exchange relaxation and partial depolarization from coated cell walls. The technique substantially enhances the spin precession quality factor and maintains a stable gyromagnetic ratio independent of spin polarization, even under frequent collisions. These findings offer a pathway for mitigating dominant relaxation channels in alkali-metal-based applications and experiments, particularly in anti-relaxation-coated cells.

quant-ph↗

arXiv2022

Massively degenerate coherent perfect absorber for arbitrary wavefronts

One of the key insights in the emerging field of non-Hermitian photonics is that well-established concepts like the laser can be operated in reverse to realize a 'coherent perfect absorber' (CPA). While conceptually appealing, such CPAs are limited so far to a single, judiciously shaped wavefront or 'mode'. Here, we demonstrate how this limitation can be overcome by time-reversing a 'degenerate cavity laser', based on a unique cavity that self-images any incident light-field onto itself. Placing a weak, critically-coupled absorber into this cavity, we demonstrate that any incoming wavefront, even a complex and dynamically-varying speckle pattern, is absorbed with close to perfect efficiency in a massively parallel interference process. Moreover, the coherent nature of multi-mode absorption allows us to tune the degree of absorption over a wide range. These characteristics open-up interesting new possibilities for applications in light-harvesting, energy delivery, light control, and imaging.

physics.optics↗

arXiv2022

Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps

Transformer-based language models significantly advanced the state-of-the-art in many linguistic tasks. As this revolution continues, the ability to explain model predictions has become a major area of interest for the NLP community. In this work, we present Gradient Self-Attention Maps (Grad-SAM) - a novel gradient-based method that analyzes self-attention units and identifies the input elements that explain the model's prediction the best. Extensive evaluations on various benchmarks show that Grad-SAM obtains significant improvements over state-of-the-art alternatives.

cs.LG↗