Undergraduate Publications

2024

Ram Dyuthi Sristi, Ofir Lindenbaum, Shira Lifshitz, Maria Lavzin, Jackie Schiller, Gal Mishne, Hadas Benisty, Contextual Feature Selection with Conditional Stochastic Gates, The Forty-first International Conference on Machine Learning, Vienna, Austria. July 2024.

abstractproject

Feature selection is a crucial tool in machine learning and is widely applied across various scientific disciplines. Traditional supervised methods generally identify a universal set of informative features for the entire population. However, feature relevance often varies with context, while the context itself may not directly

affect the outcome variable. Here, we propose a novel architecture for contextual feature selection where the subset of selected features is conditioned on the value of context variables.

Our new approach, Conditional Stochastic Gates (c-STG), models the importance of features using conditional Bernoulli variables whose parameters are predicted based on contextual variables. We introduce a hypernetwork that maps context variables to feature selection parameters to learn the context-dependent gates along with a prediction model. We further present a theoretical analysis of our model, indicating that it can improve performance and flexibility over population-level methods in complex feature selection settings. Finally, we conduct an extensive benchmark using simulated and real-world datasets across multiple domains demonstrating that c-STG can lead to improved feature selection capabilities while enhancing prediction accuracy and interpretability.

Mustafa Mahamed, Nawal Sheikh, Denis Dikarov, Direction Of Arrival (DOA) Estimation For Radars In Near-Field Regions, IEEE Conference on Microwaves, Communications, Antennas, Biomedical Engineering and Electronic Systems (COMCAS) 2024

abstractproject

The Direction of Arrival (DOA) estimation in the near-field has received significant attention in recent years due to its wide range of applications in various fields, including gesture recognition, radar systems, and acoustic imaging. Multiple DOA algorithms have been developed for far-field regions, such as Beamforming and multiple signal classification (MUSIC). These algorithms depend on far-field assumptions, such as simplifying electromagnetic waves as parallel planes. Conversely, in the near-field regions, the assumption of plane waves is not applicable; electromagnetic waves exhibit a spherical pattern. As a result, traditional DOA estimation algorithms based on far-field assumptions fail. To bridge this gap, we introduce a modification to these algorithms in order to turn them viable in the near-field regions. Instead of proposing a new DOA algorithm for the near-field region, the goal of this paper is to propose a modification to the covariance matrix used in various far-field DOA algorithms to make them applicable in the near-field regions. We first introduce the basic concepts of DOA estimation[3] and a couple of far-field DOA estimation algorithms. Then, we discuss the challenges facing these algorithms in the near-field regions. We proceed with presenting a modification to these algorithms. We finish by demonstrating the improvement through empirical evaluations and comparisons.

Neta Gevirtzer, Ofir Ben Yosef, Alon Eilam, Acoustic User Authentication with Smartphones, IEEE Conference on Microwaves, Communications, Antennas, Biomedical Engineering and Electronic Systems (COMCAS) 2024

abstractproject

This paper describes a work on an acoustic user authentication system using smartphones. The system implements two-factor authentication for Windows workstations, where the authentication procedure, including locking and unlocking the workstation is transparent to the user. Since workstations and smartphones have built-in microphones and speakers, the system does not require additional hardware. The uniqueness of the solution is being based on acoustic signals. These signals are transmitted by the user’s smartphone and received by the workstation microphone. The system is ”pure play acoustic” since no wiring or radio transmission is used. The system configuration supports multiple users in the same area. Eavesdropping prevention is provided by sequentially generated random onetime keys. Acoustic communication can be applied either in the audible range or beyond the human hearing range depending on the sampling rate of the smartphone and the workstation.

2023

Mendelman Harel, Massas Tomer and Moshe Yair, Analysis of Waves and Human Interaction Using Beach Webcams, OCEANS 2023

abstractproject

Automatic monitoring of beach video streams is important for improving safety, environmental monitoring, research, and education related to beach activities. This paper introduces a novel approach for monitoring water bodies by analyzing beach video streams. In contrast to earlier works, in the proposed approach we analyze not only the behavior of water bodies, or of humans on beach videos, but also the interactions between them. By integrating human activity and water behavior analysis, the approach provides new insights that
are unattainable by analyzing each separately. To accomplish this objective, deep neural networks are utilized to analyze video streams from existing beach webcams. The analysis includes the detection and tracking of people and waves, as well as higher level analysis. We use the task of characterizing surfing conditions as a case study and demonstrate our ability to estimate the values of key parameters that can help determine the quality of a surf spot at a given time.

2022

Yuval Ben-Hur, Asaf Goren, Da-El Klang, Yongjune Kim and Yuval Cassuto, Mitigating Noise in Ensemble Classification with Real-Valued Base Functions, IEEE International Symposium on Information Theory (ISIT) 2022

abstractproject

In data-intensive applications, it is advantageous to perform some partial processing close to the data, and communicate to a central processor the partial results instead of the data itself. When the communication medium is noisy, one must mitigate the resulting degradation in computation quality. We study this problem for the setup of binary classification performed by an ensemble of functions communicating realvalued confidence levels. We propose a noise-mitigation solution that works by optimizing the aggregation coefficients at the central processor. Toward that, we formulate a post-training gradient algorithm that minimizes the error probability given the dataset and the noise parameters. We further derive lower and upper bounds on the optimized error probability, and show
empirical results that demonstrate the enhanced performance achieved by our scheme on real data.

Tomer Fireaizen, Saar Ron and Omer Bobrowski, Alarm Sound Detection Using Topological Signal Processing, IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022

abstractproject

We present a novel approach to alarm sound detection using topological data analysis. Our main focus is on proposing a new set of robust features, based on algebraic topology, that are aimed at capturing global structural information about the dynamical system underlying each input signal. In short, we convert each signal into a point cloud and compute its corresponding persistent homology, from which we can extract a variety of useful numerical features. We demonstrate the power of this framework using the UrbanSound8K dataset and show that, by combining topological features with a classical classification method, we achieve state-of-the-art results.

Tomer Fireaizen, Gal Metzer, Dan Ben-David, Yair Moshe, Israel Cohen and Emil Bj¨ornson, Intelligent Reflecting Surface OFDM Communication with Deep Neural Prior, IEEE International Conference on Communications (ICC), 2022, awarded

abstractproject

An Intelligent Reflecting Surface (IRS) is an emerging technology for improving the data rate over wireless channels by controlling the underlying channel. In this paper, we describe a novel solution for IRS configuration to maximize the data rate over wideband channels. The optimization is obtained by online training of a deep generative neural network. Inspired by related works in image processing, this network is randomly initialized and acts as a regularization term for the optimization process since the structure of the generator is sufficient to capture a great deal of IRS statistics prior to any learning. In contrast to recent deep learning techniques for IRS configuration, the proposed technique does not require an offline training stage and can adapt quickly to any environment. Compared to the previous state-of-the-art algorithm, the proposed method is significantly faster and obtains IRS configurations that achieve higher data transmission rates.

Karin Jakoel, Liron Efraim and Tamar Rott Shaham, GANs Spatial Control via Inference-Time Adaptive Normalization, IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2022

abstractproject

We introduce a new approach for spatial control over the generation process of Generative Adversarial Networks (GANs). Our approach includes modifying the normalization scheme of a pre-trained GAN at test time, so as to act differently at different image regions, according to guidance from the user. This enables to achieve different generation effects at different locations across the image. In contrast to previous works that require either fine-tuning the model’s parameters or training an additional network, our approach
uses the pre-trained GAN as is, without any further modifications or training phase. Our method is thus completely generic and can be easily incorporated into common GAN models. We show our technique to be useful for solving a line of image manipulation tasks, allowing different generation effects across the image, while preserving the GAN’s high visual quality.

2021

Tomer Fireaizen, Dan Ben-David, Shaked Hadad, Gal Metzer, Nir Kurland, Sima Etkind, Pavel Lifshits, Yair Moshe and Israel Cohen, Intelligent Reflecting Surface Configuration Using Adaptive Quantization and Neural Prior, 8th International IEEE Conference on Microwaves, Communications, Antennas, and Electronic Systems (COMCAS) 2021

abstractproject

Intelligent Reflective Surface (IRS) is a promising technology for improving the data transmission rate in hard direct channel conditions. In this paper, we describe our solution to estimate the relevant channels and configure the IRS for efficient wireless communications, as part of the 2021 IEEE Signal Processing Cup (SP Cup) competition. First, we estimate the wireless channel and then find an IRS configuration that maximizes the rate of that channel. We begin with the provided far-from-optimal IRS configurations and apply an iterative optimization technique based on gradient descent and adaptive quantization. Further optimization is obtained by training a deep generative neural network to find a configuration that maximizes the rate function. Compared to the best provided configurations
that provide a weighted average rate of 104.07 Mbit/s, the best configurations we discovered provide a significantly higher average rate of 120.70 Mbit/s. Non-IRS based solution provides an average rate of 4.38 Mbit/s.

Yoad Pilosof, Aviv-Abraham Golan, Alon Eilam and Pavel Lifshits, Radar Targets Classification with Manifold Learning, IEEE International Conference on Microwaves, Communications, Antennas & Electronic Systemes (COMCAS), 2021

abstractproject

In this work we propose a new approach for Radar targets classification by using Manifold Learning. We focus on the acute problem of differentiating UAVs from birds.

Micro Movements of Radar target, resulting from UAVs rotating propellers or birds’ wings flapping cause scattering of the Radar echo signal in the frequency domain, named Micro Doppler. In previous works, the Micro Doppler effect was analyzed with classical methods such as SVM, KNN and Bayes. However, the

large dimensionality of the data, as well as the small dataset has limited the performance achieved. The innovation in our work is applying Manifold Learning based preprocessing of the Range-Doppler map prior to SVM classification. Our results show the effectiveness of the proposed method. With real life dataset, we have achieved 82.9% correct classification per single target hit, Vs. 74.4% achieved by classical methods on the same data set.

Dana Maklada, Yasmine Obeid-Zoabi, Alon Eilam and Michael Dikshtein, High Accuracy Distance Measurement Using Frequency Comb, IEEE International Conference on Microwaves, Communications, Antennas & Electronic Systems (COMCAS), 2021

abstractproject

High precision indoor position estimation enables new opportunities for a variety of commercial, industrial and consumer applications. In this paper, we consider a phasebased method to calculate range from noisy measurements of a Frequency Comb in a multi-fading environment. It can be used to determine the range between devices for the next-generation High Accuracy Distance Measurement (HADM) protocol. We have conducted a quantitative analysis of various estimation approaches, considering both Monte-Carlo simulations of synthetic data in a variety of ranges. Moreover, we have proposed evaluation schemes for situations in which only a subset of data is available and some information may be missing, which extends existing approaches.

2020

Zacharie Cohen, Andy Rodan, Alon Eilam and Pavel Lifshits, Acoustics Based Proximity Detector for Mobile Phones, 2020 IEEE Workshop on Signal Processing Systems (SiPS)

abstractproject

Modern mobile phones are equipped with an infrared photoelectric proximity sensor, most commonly applied to turn off the touch screen during a phone call to prevent accidental touches when users’ face/ear is detected in proximity to the screen. In this work we propose to achieve the sensing functionality without using a special sensor. Specifically, we use an already existing speaker and microphones for proximity sensing, without interfering with their originally intended operation. We build our method on the observation that the transfer function from the mobile phone speaker to the microphones varies as a function of objects located in the vicinity of the mobile phone. We have prototyped our solution on 2 mobile phones – Samsung Galaxy A70 and Xiaomi Mi 9, and achieved 99% accuracy of detection. We demonstrate that our method is robust to various environments, geometries, fabrics, and performs well in the presence of noise. We also evaluate the performance of various implementations, and discuss their trade-offs.

Oded Schlesinger, Nitai Vigderhouse, Danny Eytan and Yair Moshe, Blood Pressure Estimation From PPG Signals Using Convolutional Neural Networks And Siamese Network, 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain.

abstractproject

Blood pressure (BP) is a vital sign of the human body and an important parameter for early detection of cardiovascular diseases. It is usually measured using cuff-based devices or monitored invasively in critically-ill patients. This paper presents two techniques that enable continuous and noninvasive cuff-less BP estimation using photoplethysmography (PPG) signals with Convolutional Neural Networks (CNNs). The first technique is calibrationfree. The second technique achieves a more accurate measurement by estimating BP changes with respect to a patient’s PPG and ground truth BP values at calibration time. For this purpose, it uses Siamese network architecture. When trained and tested on the MIMIC-II database, it achieves mean absolute difference in the systolic and diastolic BP of 5.95 mmHg and 3.41 mmHg respectively. These results almost comply with the AAMI recommendation and are as accurate as the values estimated by many home BP measuring devices.

Oded Schlesinger, Nitai Vigderhouse, Yair Moshe and Danny Eytan, Estimation and Tracking of Blood Pressure Using Routinely Acquired Photoplethysmographic Signals and Deep Neural Networks, Critical Care Explorations Journal: April 2020 – Volume 2 – Issue 4 – p e0095

abstractproject

Continuous tracking of blood pressure in critically ill patients allows rapid identification of clinically important changes and helps guide treatment. Classically, such tracking requires invasive monitoring with its associated risks, discomfort, and low availability outside critical care units. We hypothesized that information contained in a prevalent noninvasively acquired signal (photoplethysmograph: a byproduct of pulse oximetry) combined with advanced machine learning will allow continuous estimation of the patient’s blood pressure.

2019

Guy Dascalu, Omer Movshovits and Alon Eilam, Pure Play Ultrasonic 3D Positioning System with Unsynchronized Beacons and Receivers, IEEE COMCAS 2019 International Conference on Microwaves, Communications, Antennas & Electronic Systems, Tel-Aviv, Israel

abstractproject

This paper describes a work on 3D ultrasonic positioning system for mobile devices such as cellular phones and IoT devices with built-in microphones. The uniqueness of the described system is its ability to provide simultaneous and accurate positioning of multiple devices, based on ultrasonic signals transmitted by stationary beacons, while the devices are sampling the signals by their internal, unsynchronized clocks. Auxiliary information required for synchronization is collected and broadcasted by a monitor unit through an ultrasonic modem. The system is considered to be “pure play ultrasonic” since no wiring or radio transmission is used. The number of devices positioned simultaneously is practically unlimited since they do not emit any energy, and their position calculation is performed locally. In the demonstrated system the positioned devices were cellular phones, operating at a sampling rate of 44,100 samples per second. This sampling rate enables using ultrasonic signals beyond the human hearing range.

Adar Elad, Doron Haviv, Yochai Blau and Tomer Michaeli, Direct Validation of the Information Bottleneck Principle for Deep Nets, ICCV 2019 Workshop on Statistical Deep Learning in Computer Vision, Seoul, Korea.

abstractproject

The information bottleneck (IB) has been suggested as a fundamental principle governing performance in deep neural nets (DNNs). This idea sparked research on the information plane dynamics during training with the cross-entropy loss, and on using the IB of some “bottleneck” layer as a loss function. However, the claim that reaching the maximal value of the IB Lagrangian in each layer leads to optimal performance, was in fact never directly confirmed. In this paper, we propose a direct way of validating this hypothesis, using layer-by-layer training with the IB loss. In accordance with the original theory, we train each DNN layer explicitly with the IB objective (and without any classification loss), and freeze it before moving on to train the next layer. While mutual information (MI) is generally hard to estimate in high dimensions, we show that in the case of MI between DNN layers, this can be done quite accurately using a modification of the recently proposed mutual information neural estimator [4]. Interestingly, we find that layer-by-layer training with the IB loss leads to accuracy which is on-par with end-to-end training with the cross entropy loss. This is, thus, the first direct experimental illustration of the link between the IB value in each layer, and a net’s performance.

Yoel Bud, Guy Shkury, Roee Diamant, Yotam Zuriel and Aviad Scheinin, Robust Automatic Detector And Feature Extractor For Dolphin Whistles, IEEE Oceans 2019, Marseille France.

abstractproject

A key in Dolphin’s conservation efforts is population estimation in their natural environment. A common method for mapping Dolphin’s appearance is the detection of their vocalizations. In this paper, we propose a novel detection technique for Dolphin’s whistles, referred to as ECV (Entropy, Correlation, and Viterbi algorithm). ECV is a robust detector of low complexity that automatically detects dolphin’s whistles and extracts their spectral features, using a single receiver with only a few system parameters. The method employs a chain of decisions based on spectral entropy and time-domain correlation followed by constrained Viterbi algorithm to extract the whistles’ features. Simulation results as well as performance over real recordings shows a good trade off between detection and false alarm, that compares well with the widely used PAMguard system.

Gal Maman, Or Yair, Danny Eytan and Ronen Talmon, Domain Adaptation Using Riemannian Geometry of SPD Matrices, 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK.

abstract

In this paper, we propose a new unsupervised domain adaptation method based on the Riemannian geometry of Symmetric PositiveDefinite (SPD) matrices. The proposed domain adaptation is based on parallel transport (PT) and moments alignment. We show that this method facilitates meaningful comparisons between data points from different domains, while preserving the inherent internal structure of each domain. Experimental results demonstrate the adaptation of high-dimensional noisy electrophysiological signals collected from different subjects.

Kobi Shiran, Gal Kinberg, Or Yair, Yair Moshe and Ronen Talmon, Manifold Learning for Data-driven Dynamical System Modeling, 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) Show & Tell, Brighton, UK.

abstractproject

The extraction of models from data (in a sense, the “understanding” of the physical laws giving rise to the data) is a fundamental cognitive as well as a scientific challenge. The demonstration we present revolves around a geometric/analytic learning approach capable of creating minimal descriptions of parametrically dependent unknown nonlinear dynamical systems. This is accomplished by a data-driven extraction of useful intrinsic-state variables and parameters, in terms of which, one can empirically model the underlying dynamics. This approach follows recent trends in data analysis and signal processing, operating directly on observations, systematically creating accurate representations from data, without deriving models in closed-form and without any prior knowledge about the system dynamics. In particular, we present a kernel-based manifold learning approach, which learns the intrinsic geometric structure underlying the observations by capturing and exploiting the co-dependencies between the different dimensions of the data.

2018

Samah Khawaled, Mohamad Khateeb and Hadas Benisty, Audio Retrieval By Voice Imitation, 2018 International Conference on the Science of Electrical Engineering (ICSEE), Eilat, Israel, best student paper award

abstractproject

Existing sound retrieval systems are mostly based on a textual query. Using text to describe a sound signal is not intuitive and is often inaccurate due to subjective impression of the user; different people may use different words to describe the same sound which makes theses system complex to design and unintuitive to use. Vocal imitation, however, is the most natural human way to describe a sound. In this paper we consider a newly rising approach for sound retrieval based on vocal imitations, where the user records himself imitating the desired sound, and the system retrieves a ranked list of the most similar sounds in the dataset. In this work we represent sound signals using histograms, obtained with respect to a Gaussian Mixture Model (GMM), representing the spectral domain. This recently proposed approach was successfully applied for word representation in a keyword spotting task. Having a fixed length representation for
vocal imitation signals allows us to train a robust classifier using support vector machine (SVM). Given a test imitation signal, we
apply the classifier and use the output score to rank the retrieved signals, based on a majority vote. Our simulation results show that
the proposed system yields a more accurate ranking compared with other existing solutions.

Guy Feferman, Michal Blatt and Alon Eilam, Indoor Positioning with Unsynchronized Sound Sources, 2018 International Conference on the Science of Electrical Engineering (ICSEE), Eilat, Israel.

abstractproject

This paper describes a work on indoor positioning system for cellular phones and IoT devices. The uniqueness of the system is its ability to provide simultaneous and accurate positioning of many users, based on merely three acoustic signals transmissions from stationary Beacons, without timing synchronization or position calibration. In the demonstrated system, cellular phones were operating with their internal, unsynchronized clock at a sampling rate of 44,100 samples per second. This sampling rate has enabled using positioning signals beyond the human hearing range, not interfering with audible signals. The system was deployed and demonstrated successfully. in crowded places while providing positioning accuracy of centimeters.

Ifat Abramovich, Tomer Ben-Yehuda and Rami Cohen, Low-Complexity Video Classification using Recurrent Neural Networks Extended Abstract, 2018 International Conference on the Science of Electrical Engineering (ICSEE), Eilat, Israel.

abstractproject

YouTube-8M [1] is a large-scale video dataset containing more than 8 million labeled YouTube videos. Announced by Google in September 2016, it is the first video classification dataset whose size and diversity are comparable to existing image datasets. The labels of the videos in the dataset are machine-generated, from a vocabulary of 3862 categories. A video may have multiple labels, where on average there are 3 labels per video. Although YouTube-8M dataset has a great potential and had already advanced research on video classification, there are still difficulties arising from computational challenges related to the size of the dataset, from noisy and incomplete labels, and from higher complexity compared to image datasets due to the temporal dimension.

In this work, we train several recurrent neural networks for video classification using a subset of YouTube-8M dataset. We focus on computational complexity analysis, and provide experimental results aimed at enabling the efficient development of low-complexity network architectures for both fast training and inference. Similarly to [1], our approach is based on extracting frame-level features using the powerful Inception-V3 architecture. We use recurrent neural networks with either long short-term memory (LSTM) or bidirectional LSTM layers to capture temporal information. Finally, we evaluate the performance of a transfer-learning approach using an architecture trained on only a part of our dataset. We demonstrate that transfer learning offers a good complexity performance tradeoff for the video classification task.

Ron Dorfman, Etai Wagner, Almog Lahav, Alon Amar, Ronen Talmon and Yaron Halle, Spatio-Temporal Detection of Cumulonimbus Clouds in Infrared Satellite Images, 2018 International Conference on the Science of Electrical Engineering (ICSEE), Eilat, Israel., Best Student Paper Award

abstractproject

In this paper, we address the problem of Cumulonimbus (Cb) cloud detection from Infrared (IR) satellite images. The detection of such storm clouds is of high importance since they pose extreme danger to aviation. We present a joint spatiotemporal detection method that exploits the distinct spatial characteristics of Cb clouds as well as their prototypical evolution over time. The presented method is unsupervised and does not require labeled data or predefined spatial handcrafted features, such as particular shapes, temperatures, textures, and gradients. We demonstrate the performance of the proposed method on several sequences of IR satellite images taken from the middle east region.

Roni Ash, Dolev Ofri, Jonathan Brokman, Idan Friedman and Yair Moshe, Real-time Pedestrian Traffic Light Detection, 2018 International Conference on the Science of Electrical Engineering (ICSEE), Eilat, Israel.

abstractproject

rossing a road is a dangerous activity for pedestrians and therefore pedestrian crossings and intersections often include pedestrian-directed traffic lights. These traffic lights may be accompanied by audio signals to aid the visually impaired. In many cases, when such an audio signal is not available, a visually impaired pedestrian cannot cross the road without help. In this paper, we propose a technique that may help visually impaired people by detecting pedestrian traffic lights and their state (walk/don’t walk) from video taken with a mobile phone camera. The proposed technique consists of two main modules – an object detector that uses a deep convolutional network, and a decision module. We investigate two variants for object detection (Faster R-CNN combined with a KCF tracker, or Tiny YOLOv2) and compare them. For better robustness, we exploit the fact that abrupt switching from red to green or vice versa is unique to traffic lights. The proposed technique aims to operate on a mobile phone in a client-server architecture. It proves to be fast and accurate with running time of 6 ms per frame on a desktop computer with GeForce GTX 1080 GPU and detection accuracy of more than 99%.

Guy Feferman, Guy Shofen and Alon Eilam, Acoustic Positioning with Unsynchronized Sound Sources, 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) Show & Tell, Calgary, Alberta, Canada., In collaboration with SONARAX

project

Sharon Israely, Gerry Leisman, Chay C. Machluf and Eli Carmeli, Muscle Synergies Control during Hand-Reaching Tasks in Multiple Directions Post-stroke, Frontiers in Computational Neuroscience, 12 (10), 2018.

project

2017

Sharon Israely, Gerry Leisman, Chay Machluf, Tal Shnitzer and Eli Carmeli, Direction Modulation of Muscle Synergies in a Hand-Reaching Task, IEEE Transactions on Neural Systems and Rehabilitation Engineering, 25 (12), 2017

abstractproject

Functional tasks of the upper extremity can be executed by a variety of muscular patterns, independent of the direction, speed and load of the task. This large number of degrees of freedom imposes a significant control burden on the CNS. Previous studies suggested
that the human cortex synchronizes a discrete number of neural functional units within the brainstem and spinal cord, i.e. muscle synergies, by linearly combining them to execute a great repertoire of movements. Further exploring this control mechanism, we aim to study whether a single set of muscle synergies might be generalized to express movements in different directions. This was implemented by using a modified version of the non-negative matrix factorization algorithm on EMG data sets of the upper extremity of healthy people. Our twelve participants executed hand-reaching movements in multiple directions. Muscle synergies that were extracted from movements to the center of the reaching space could be generalized to synergies for other movement directions. This finding was
also supported by the application of a weighted correlation matrix, the similarity index and the results of the K-means cluster analysis. This might reinforce the notion that the CNS flexibly combines a single set of small number of synergies in different amplitudes to modulate movement for different directions.

Dean Carmel, Ariel Yeshurun and Yair Moshe, Detection of Alarm Sounds in Noisy Environments, 25th European Signal Processing Conference (EUSIPCO) 2017, Kos, Greece.

abstractproject

Sirens and alarms play an important role in everyday life since they warn people of hazardous situations, even when these are out of sight. Automatic detection of this class of sounds can help hearing impaired or distracted people, e.g., on the road, and contribute to their independence and safety. In this paper, we present a technique for the detection of alarm sounds in noisy environments. The technique is not limited to particular alarms and can detect most electronically generated alerting sounds within 200 ms. We consider a set of acoustic features and use the ReliefF algorithm to select only the ones that best differentiate between alarms and other sounds. We use an SVM classifier as the detector. On the tested dataset, consisting of several dozen alarm sounds and several dozen background noises, the proposed technique shows an accuracy of 98% per audio frame. With a larger training dataset, this result is expected to substantially improve.

Nadav Bhonker, Shai Rozenberg and Itay Hubara, Playing SNES In The Retro Learning Environment, 5th International Conference on Learning Representations (ICLR) 2017, Workshop track , Toulon, France.

abstractproject

Mastering a video game requires skill, tactics and strategy. While these attributes may be acquired naturally by human players, teaching them to a computer program is a far more challenging task. In recent years, extensive research was carried out in the field of reinforcement learning and numerous algorithms were introduced, aiming to learn how to perform human tasks such as playing video games. As a result, the Arcade Learning Environment (ALE) (Bellemare et al., 2013) has become a commonly used benchmark environment allowing algorithms to train on various Atari 2600 games. In many games the state-of-the-art algorithms outperform humans. In this paper we introduce a new learning environment, the Retro Learning Environment — RLE, that can run games on the Super Nintendo Entertainment System (SNES), Sega Genesis and several other gaming consoles. The environment is expandable, allowing for more video games and consoles to be easily added to the environment, while maintaining a simple unified interface. Moreover, RLE is compatible with Python and Torch. SNES games pose a significant challenge to current algorithms due to their higher level of complexity and versatility.

2016

Gal Binyamin, Alon Eilam and Itai Dagan, Delivery of QR Codes to Cellular Phones through Data Embedding in Audio, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

We describe a system that delivers a website address to a cellular phone by encoding inaudible binary data in an analogue audio signal, which is received by the microphone of the cellular phone. This is an alternative to encoding a web site address in a QR code label, which is scanned by the cellular phones camera. Data embedding in the audio signal is done by modifying the phase of the signal’s modulated complex lapped transform (MCLT) coefficients, while the perceived quality of the embedded audio signal remains the same as that of the original audio signal. A whole system was implemented and tested both in simulation and in reality. The data rate achieved in reality at a distance of 1.5 meters was 48 bits per second. This rate is sufficient to deliver the same amount of information contained in a QR code label faster and easier than by camera scan.

Amit Oren, Avi Matzliach, Rami Cohen and Hagit Friedman, Cry-based Detection of Developmental Disorders in Infants, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

Developmental disorders are a group of neurological conditions originating at childhood, which involve serious impairments in language, learning, and motor skills. An early detection of developmental disorders is crucial, as it enables early intervention (e.g., speech-language and occupational therapy)
that may reduce neurological and functional deficits. In this work, we develop a tool for an early identification of developmental disorders in infants based on their cry. We use signal-processing tools to extract distinguishing cry features (e.g., pitch and formants), and exploit their correlation with the risk of having developmental disorders. To estimate this risk, we train a k-NN machine-learning system. Performance is evaluated against a database of diagnosed infants, with 89% accuracy in cross validation testing.

Ori Shahar, Lee Twito and Nurit Spingarn, Speaker Diarization Using Locally Linear Embbeding Dimension Reduction, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

Speaker diarization is a significant part of many applications in today’s fast growing user-end software and technologies. In the last decade, speaker diarization has attracted significant research effort, however most of the speaker diarization methods deeply rely on statistical models which become unreliable in case of short utterances diarization and in noisy conditions. In this paper, we introduce a speaker diarization system which is based on the Locally Linear Embedding (LLE). The LLE enables to extract the inherent structure of the data and thus provide better clustering. Experimental results show error rates lower than 10% and improved stability in comparison with a conventional speaker diarization method.

Yehav Alkaher, Osher Dahan and Yair Moshe, Detection of Distress in Speech, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

A distress situation is reflected in a person’s speech e.g., fear, or in the speech of those around him e.g., anger. For many real-life situations, it may be desired to detect distress by remote monitoring. This paper deals with such remote monitoring using a microphone. Namely, we propose a technique for speaker-independent detection of distress in speech. Different temporal and spectral acoustic features such as the Mel frequency cepstral coefficients (MFCC) and the Teager energy operator (TEO) are investigated and the most relevant ones for the task are selected using the ReliefF feature selection algorithm. We use these features to train an SVM classifier in order to differentiate between speech in a distress situation and speech in which no distress is presented. On the Berlin Database of Emotional Speech, the proposed technique achieves classification accuracies of 91% per utterance and 87.8% per time window.

Amir Ivry, Talor Abramovich and Hadas Benisty, Enhancement Of BCI Classifiers Through Domain Adaptation, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

Clinical Brain-Computer Interface (BCI) systems seek to enable paralyzed individuals to operate devices with their brain activity. Non-invasive systems based on electroencephalographic (EEG) signals are popular since they avoid risks associated with invasive procedures, but unfortunately EEG signals are inherently noisy, making effective classifiers challenging to develop. Commonly, new classifiers are benchmarked on signals from healthy subjects executing physical movements, under the assumption that the performance will transfer to clinical cases where only imagined movements are possible. Here, we show in contrast that classifiers trained on signals associated with actual movements perform erratically when applied to signals associated with imagined movements. We suggest that this is because the signals lay in different domains. Then, to exploit the different statistical distributions, we apply a domain adaptation technique, Frustratingly Easy Domain Adaptation (FEDA), improving classifier performance accuracy by a third on a discrimination task that simulates the clinical condition.

Maya Rapaport, Noga Cohen, Tal Shnitzer, Natalya Yarovinsky(*), Ronen Talmon and Judith Aharon-Peretz(*), Alternating Diffusion Maps for Dementia Severity Assessment, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel., (*) Rambam Health Care Campus, Technion, Israel Institute of Technology

abstract

In this paper we address the detection of Alzheimer’s disease based solely on EEG recordings. We assume that the state of Alzheimer’s disease can be described by a latent manifold, captured by the EEG sensors and apply alternating diffusion to reveal this common underlying manifold from multiple EEG sensors. We show that based on a small number of EEG electrodes, a new representation can be obtained, which allows a clear distinction between healthy subjects and Alzheimer patients in different disease stages.

Ran Gladstone, Avihai Barel, Elior Shenhav and Yair Moshe, Distance Estimation for Marine Vehicles Using a Monocular Video Camera, 24th European Signal Processing Conference (EUSIPCO 2016), Budapest, Hungary.

abstractproject

Distance estimation for marine vessels is of vital importance to unmanned ship vehicles (USVs) for navigation and collision prevention. This can be achieved by means such as radar or laser sighting. However, due to constraints of the USV, it may be desired to estimate distance using a monocular camera. In this paper, we propose a method that, given a video of a marine vehicle in a maritime environment and a tracker, estimates the distance of the tracked vehicle from the camera. The method detects the horizon and uses its distance as a reference. It detects the contact point of the vehicle with the sea surface by finding a maximally stable extremal region (MSER). Then, it relies on geometries of the earth and on optical properties of the camera to compute the distance. The method was tested on video footage of several sea maneuvers with an average error of 7.1%.

Yuval E. Yaish, Yair Moshe, Maor Itzhak, and et al., Automated Circuit Fabrication and Direct Characterization of Carbon Nanotube Vibrations, Nature Communications, 7 (2016): 12153.

abstract

Since their discovery, carbon nanotubes have fascinated many researchers due to their unprecedented properties. However, a major drawback in utilizing carbon nanotubes for practical applications is the difficulty in positioning or growing them at specific locations. Here we present a simple, rapid, non-invasive and scalable technique that enables optical imaging of carbon nanotubes. The carbon nanotube scaffold serves as a seed for nucleation and growth of small size, optically visible nanocrystals. After imaging the molecules can be removed completely, leaving the surface intact, and thus the carbon nanotube electrical and mechanical properties are preserved. The successful and robust optical imaging allowed us to develop a dedicated image processing algorithm through which we are able to demonstrate a fully automated circuit design resulting in field effect transistors and inverters. Moreover, we demonstrate that this imaging method allows not only to locate carbon nanotubes but also, as in the case of suspended ones, to study their dynamic mechanical motion.

Amnon Balanov, Arik Schwartz and Yair Moshe, Reduced-Reference Image Quality Assessment Based on DCT Subband Similarity, 8th International Conference on Quality of Multimedia Experience ( QoMEX 2016), Lisbon, Portugal.

abstract

Reduced-reference image quality measures aim to estimate the visual quality of a distorted image with only partial information about the “perfect quality” reference image. In this paper, we present a reduced-reference image quality assessment (IQA) metric based on DCT Subbands Similarity (RR-DSS). According to the assumption that human visual perception is adapted for extracting structural information, the proposed technique measures change in structural information in subbands in the discrete cosine transform (DCT) domain and weights the quality estimates for these subbands. RR-DSS is simple to implement, incurs low computational complexity, and has a flexible tradeoff between the amount of side information and image quality estimation accuracy. RR-DSS was tested with public image databases and shows excellent correlation with human judgments of quality. It outperforms state-of-the-art RR IQA techniques and even several FR IQA techniques.

Liron Anavi, Avi Giterman, Maya Fainshtein, Vladi Solomon and Yair Moshe, A Highly Parallel Coding Unit Size Selection For HEVC, IEEE International Conference for Acoustics, Speech and Signal Processing (ICASSP 2016), Shanghai, China.

abstractproject

The High Efficiency Video Coding (HEVC) standard provides a substantial improvement in coding efficiency over previous video coding standards at the cost of a higher computational complexity. HEVC employs a quadtree based image structure by partitioning the image into coding units (CUs). Finding the optimal CU size in terms of rate-distortion is one of the most computationally challenging parts of any HEVC encoder. Previous works for fast CU size selection are usually based on data dependency between neighboring CUs and therefore limit the degree of possible parallelism. In this paper, we present a fast CU size selection method that does not depend on any data from other CUs in the same frame, thus allowing utilization of the high parallel processing capability of many-core processors, such as a GPU. Experimental results show that the proposed method incurs only a negligible loss in rate-distortion performance compared with counterpart methods that limit parallelism.

Or Dicker , Aviv Peleg , Tal Shnitzer , Oscar Lichtenstein , Yair Moshe, Real Time Control of Hand Prosthesis Using EMG, International Conference on the Science of Electrical Engineering (ICSEE 2016), Eilat Israel.

abstractproject

Current solutions for below the elbow amputees include affordable prosthesis allowing only a single movement or highly expensive prosthesis allowing several gestures. In this project, our goal was to design a system that provides an inexpensive, multi-functional solution for the hand prosthesis problem. We construct a real-time, portable system based on the Myo armband and a 3D printed prosthesis and show that this framework can provide a good and inexpensive solution for below the elbow amputees of all ages.

2015

Amnon Balanov, Arik Schwartz, Yair Moshe and Nimrod Peleg, Image Quality Assessment Based on DCT Subband Similarity, IEEE International Conference on Image Processing (ICIP 2015), Quebec, Canada.

abstractproject

Measuring image quality becomes increasingly important due to the many applications involving digital imaging and communication. Image quality assessment aims to develop a visual quality metric that correlates well with human visual perception. In this paper, we present a full-reference image quality assessment technique based on DCT Subbands Simi-larity (DSS). The proposed technique exploits important characteristics of human visual perception by measuring change in structural information in subbands in the discrete cosine transform (DCT) domain and weighting the quality es-timates for these subbands. The proposed technique was tested with public image datasets and shows higher correla-tion with subjective results than state-of-the-art techniques. Another advantage of the proposed technique is its low com-putational cost.

2014

Tamar Rott, Dorin Shriki and Tamir Bendory, Edge Preserving Multi-Modal Registration Based On Gradient Intensity Self-Similarity, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstract

Image registration is a challenging task in the world of medical imaging. Particularly, accurate edge registration plays a central role in a variety of clinical conditions. The Modality Independent Neighbourhood Descriptor (MIND) demonstrates state of the art alignment, based on the image self-similarity. However, this method appears to be less accurate regarding edge registration. In this work, we propose a new registration method, incorporating gradient intensity and MIND self-similarity metric. Experimental results show the superiority of this method in edge registration tasks, while preserving the original MIND performance for other image features and textures.

Aviv Barabi, Dvir Sason and Rami Cohen, Low Complexity Image Compression of Capsule Endoscopy Images, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstractproject

Capsule endoscopy is a method for recording images of the digestive tract. A patient swallows a capsule containing a tiny camera, which captures images that are then transmitted wirelessly to an external receiver for examination by a physician. The images are captured using a Bayer filter mosaic, such that each pixel in raw captured images represents only one color: red, green or blue. Due to limited computational capabilities in the capsule and bandwidth constraints, low-complexity and efficient compression of Bayer endoscopy images is required before transmission. In this paper, we focus on the JPEG algorithm, and show how to adapt it for compressing Bayer images. We show that by transformation to the YCgCo color space and appropriate optimization of parameters, significant improvement is achieved over the standard JPEG.

Maya Harel and Yair Moshe, Tone Mapping for Shortwave Infrared Face Images, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstractproject

Sensing in the Shortwave Infrared (SWIR) range has only recently been made practical. The SWIR band is not visible to the human eye but shows shadows and contrast in its imagery. Moreover, SWIR sensors are highly tolerant to challenging atmospheric conditions such as fog and smoke. However, fundamental differences exist in the appearance between images sensed in visible and SWIR bands. In particular, human faces in SWIR images do not match human intuition and make it difficult to recognize familiar faces by looking at such images. In this paper, we deal with a novel tone mapping application for SWIR face images. We propose a technique to map the tones of a human face acquired in the SWIR band to make it more similar to its appearance in the visible band. The proposed technique is easy to implement and produces natural looking face images.

Ben Dror, Eilon Yanai, Alex Frid, Nadav Goldenthal, Ilana Shlezinger, Shmuel Raz, Hagit Hel-Or and Nimrod Peleg, Automatic Assessment of Parkinson’s Disease From Natural Hands Movements Using 3D Depth Sensor, 28th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstractproject

Parkinson’s Disease (PD) is a degenerative disease of the central nervous system with a profound effect on the motor system. Symptoms include slowness of movement, rigidity of motion and in some patients, tremor. The severity of the disease is quantified using the Unified Parkinson Disease Rating Scale (UPDRS) which is a subjective scale performed and scored by physicians. In this work, we present an automated, objective quantitative analysis of four UPDRS motor examinations of Hand Movement and Finger Taps. For this purpose, a non-invasive system for recording and analysis of fine motor skills of hands was developed. The system is based on a simple low-cost depth acquisition sensor, similar to the second generation of Microsoft’s Kinect sensor, and novel recursive self-correcting hand tracking algorithm. The system allows patients to perform test tasks in a natural and unhindered manner. The evaluation of the system was carried out on PD patients and controls. Machine Learning based classification was performed on the acquired data, followed by a decision making scheme.

Pavel Lifshits, Alon Eilam, Yair Moshe and Nimrod Peleg, DSP in Heterogeneous Multicore Embedded Systems – A Laboratory Experiment, 22nd European Signal Processing Conference (EUSIPCO 2014), Lisbon, Portugal.

abstract

Undergraduate engineering students who are learning Digital Signal Processing (DSP) are expected to have the ability to implement their theoretical knowledge in various applications soon after graduation. In this paper, we present a laboratory experiment developed for undergraduate students that addresses the challenge of getting them familiar with implementing DSP algorithms in heterogeneous multicore systems. In a top-down approach, the students first gain control of the development environment, and then implement DSP algorithms on a general purpose and on a digital signal processor core. Through the experiment, they get to appreciate the advantages of DSP core architecture in performing signal processing algorithms, and learn methods for timing and data transfer between cores while meeting real-time constraints. In a limited time frame, this hands-on laboratory experiment exposes the students to state-of-the-art multicore development practices and increases their knowledge and interest in DSP and in embedded programming.

Gilad Avrashi, Shlomi Museri, Yaakov Bucris, Azriel Sinai and Alon Amar, Underwater Video Streaming Using Adaptive Frame Decimation, MTS/IEEE OCEANS 2014, Taipei, Taiwan.

abstractproject

We introduce a low data rate video compression scheme for underwater acoustic communication. Recently, multi-carrier based underwater acoustic modems have been proposed and tested for data rates of up to tens of kilobits-per-second for about 1-4 kilometers transmission range. Online transmission of video data using these rates requires extreme compression. Herein, we propose pre-encoder and post-decoder processing in order to reduce the data rate. This is done by exploiting the characteristics of underwater videos for excluding frames. The algorithm was tested and analyzed in simulation environment with several video samples taken by divers in shallow waters of the Mediterranean. We show that using this technique, increased performance is achieved compared to standard H.264/AVC based codecs in both objective and subjective terms.

Michael Birk, Amir Burshtein, Tanya Chernyakova, Alon Eilam, Jung Woo Choe, Amin Nikoozadeh, Pierre Khuri-Yakub and Yonina C. Eldar, Compressed 3D Ultrasound Imaging with 2D Arrays, IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2014), Florence, Italy.

abstractproject

Contemporary sonography is performed by digitally beamforming signals sampled by several transducer elements placed upon an array. High-resolution digital beamforming introduces the demand for a sampling rate significantly higher than the signal’s Nyquist rate, which greatly increases the volume of data that must be processed. In 3D ultrasound imaging, 2D transducer arrays rather than 1D arrays are used, and more scan-lines are needed for volumetric imaging. This implies that the amount of sampled data is vastly increased with respect to 2D imaging. In this work we show that a considerable reduction both in sampling rate and processing time can be achieved by applying the ideas of Xampling and frequency domain beamforming, leading to a sub-Nyquist sampling rate. We extend previous work on frequency domain beamforming for 2D ultrasound imaging to the geometry imposed by 3D tissues and a grid of transducer elements. This method uses only a portion of the bandwidth of the ultrasound signals to reconstruct the image. We demonstrate our results by imaging a phantom comprised of fishing wires, and show that by performing 3D beamforming in the frequency domain, a sub-Nyquist sampling rate and a low processing rate are obtained, while keeping adequate image quality.

2012

Tomer Hamam, Yedidyah Dordek and Deborah Cohen, Single-Band Infrared Texture-Based Image Colorization, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstract

Infrared (IR) imaging has a wide variety of applications such as night vision detection and tracking, meteorology radiometers and spectroscopy techniques. An infrared image is a monochrome image, usually presented in grayscale. It has been proved that colorization of IR images can reduce human error and speed up reaction time. Most previously suggested coloring methods use a reference color image, whose characteristic features differ drastically from IR ones. These differences make those features less pertinent to the coloring process. In this paper, we present a novel texture-based method for automatically coloring IR images. The method uses a reference (source) color image, which is selected from a database, built in advance containing various natural scenes. The source image is selected using a texture-matching algorithm that searches for a resemblance to the IR (target) image. The source and target images are divided into texture-based segments and a color segment best match is found for every IR segment. The coloring process is performed for each pair of IR-color segments, and exploits global as well as local features. Results show that our method produces more natural-looking images than achieved heretofore.

Victoria Evelkin and Israel Cohen, Effect of Latency Time in High Frequencies on Sound Localization, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstract

Tactical military headsets allow hearing protection and communication between soldiers using surround technology. In this paper, we investigate the effect of latency time between low and high frequencies on the ability of a person to correctly identify an external sound source direction with tactical military headsets. The latency time between low and high frequencies results from a digital processing unit. Low frequencies from an external sound source are not processed; therefore, they are received in the ear canal before the digitally processed higher frequencies. Two experiments were conducted using non-individualized Head Related Transfer Function and headphones for trained and untrained volunteers. The experiments were done for two types of sources: hu-man speech and white Gaussian noise, and were performed with latency times of zero, 20msec, and 40msec applied to frequencies below 20Hz. The experimental results show that the errors in sound localization accuracy in both experiments are fewer for signals without latency time, compared to processed signals with latency times between low and high frequencies.

Rami Cohen and Yizhar Lavner, Infant Cry Analysis and Detection, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstractproject

In this paper we propose an algorithm for automatic detection of an infant cry. A particular application of this algorithm is the identification of a physical danger to babies, such as situations in which parents leave their children in vehicles. The proposed algorithm is based on two main stages. The first stage involves feature extraction, in which pitch related parameters, MFC (mel-frequency cepstrum) coefficients and short-time energy parameters are extracted from the signal. In the second stage, the signal is classified using the k-NN algorithm and is later verified as a cry signal, based on the pitch and harmonics information. In order to evaluate the performance of the algorithm in real world scenarios, we checked the robustness of the algorithm in the presence of several types of noise, and especially noises such as car horns and car engines that are likely to be present in vehicles. In addition, we addressed real time and low complexity demands during the development of the algorithm. In particular, we used a voice activity detector, which disabled the operation of the algorithm when voice activity was not present. A data base of baby cry signals was used for performance evaluation. The results showed good performance of the proposed algorithm, even at low SNR.

Tom Zohar, Ido Ariav and Meir Bar-Zohar, Robust and Efficient Change Detection Algorithm based on 3D Line Segments, 27th IEEE Convention of Electrical and Electronics Engineers in Israel 2012, Eilat, Israel.

abstractproject

In this paper we examine and improve a new approach for change detection which is based on the appearance and disappearance of 3D line segments as seen in a new image. These 3D line segments are estimated from a set of learning images taken from arbitrary viewpoints and under arbitrary light conditions in an unsupervised manner. The main advantage of the proposed method lies in the fact that the change detection is performed by comparing line segments, and not surfaces or gray levels. Computing 3D surfaces in an image can be computationally intensive, and other methods such as image subtraction or cross-correlation are sensitive to lighting conditions and changes in viewpoints. Moreover, most man-made objects such as buildings, cars, and even cities viewed from above consist mainly of straight lines, and therefore this method is highly applicable for such structures. The proposed algorithm first focuses on the reconstruction of a set of 3D line segments forming a certain 3D scene using a set of 2D line segments obtained from the learning images in an unsupervised manner, without any prior knowledge on the cameras’ positions or relative distance. In the change detection stage, we use the reconstructed 3D scene of line segments to check if changes, such as appearance or disappearance of objects, have occurred in a given test image. This test image can be taken from arbitrary viewpoint and under arbitrary lighting conditions. Our change detection algorithm not only distinguishes between the states of “changed” and “not-changed” line segments, it also classifies the “changed” line segments as appeared – objects that entered the scene in the test image, and disappeared – objects that left the 3D scene reconstructed from the lines of the learning images.

Ori Bryt, Asaf Elron, Pavel Lifshits, Tsahee Zidenberg, Yair Moshe and Nimrod Peleg, A Laboratory Experiment for Real-time Echo Cancellation Using a Beagleboard, 5th IEEE European DSP Education and Research Conference (EDERC 2012), Amsterdam, Netherlands.

abstract

Practical experience is an important aspect of the training of every engineer. One way to develop such experience is by hands-on laboratory experimentation which involves cutting-edge technology. In this paper, we present a signal processing lab experiment developed for undergraduate students. Throughout the experiment, students first learn the basics of adaptive filtering and implement an adaptive echo cancellation algorithm using MATLAB. Then, students use a software framework for implementing the same echo cancellation algorithm on a BeagleBoard development platform using both ARM and fixed-point DSP cores of an embedded system-on-chip from Texas Instruments. In a relatively short amount of time (8 hours at the lab and a few additional hours of homework), students learn an important signal processing technique, as well as use complex state-of-the-art DSP hardware to implement it as part of an application running in real-time.

Elad Bullkich, Idan Ilan, Yair Moshe, Yacov Hel-Or and Hagit Hel-Or, Moving Shadow Detection by Nonlinear Tone-Mapping, 19th International Conference on Systems, Signals and Image Processing (IWSSIP 2012), Vienna, Austria.

abstractproject

Shadow detection is a well-known problem in image and video processing. Detecting moving shadows is useful in numerous applications such as object detection and tracking. Most works in this area are not suitable for shadow detection using a low-cost outdoor surveillance camera. In this work we suggest a fast shadow detection approach for video surveillance, by comparing each video frame to a continuously updated background image. We differentiate shadow areas from foreground areas by assuming the shad-ow pixels are associated with background pixels through a nonlinear tone mapping. This assumption is general and applies to various systems and scene conditions. A distance measure between patch images that account for nonlinear tone mapping is calculated by adapting a recently suggested approach for pattern matching termed Matching by Tone Mapping (MTM). We show that the proposed technique is computationally efficient and outperforms state-of-the-art shadow detection techniques in typical surveillance scenarios.

Inbal Horev, Ori Bryt and Ron Rubinstein, Adaptive Image Compression Using Sparse Dictionaries, 19th International Conference on Systems, Signals and Image Processing (IWSSIP 2012), Vienna, Austria.

abstractproject

Transform coding is a widely used image compression technique, where entropy reduction can be achieved by decomposing the image over a dictionary which provides compaction. Existing algorithms, such as JPEG and JPEG2000, utilize fixed dictionaries which are shared by the encoder and decoder. Recently, works utilizing content-specific dictionaries show promising results by focusing on specific classes of images and using highly specialized dictionaries. However, such approaches lose the ability to compress arbitrary images. In this paper we propose an input-adaptive compression approach, which encodes each input image over a dictionary specifically trained for it. The scheme is based on the sparse dictionary structure, whose compact representation allows relatively low-cost transmission of the dictionary along with the compressed data. In this way, the process achieves both adaptivity and generality. Our results show that although this method involves transmitting the dictionary, it remains competitive with the JPEG and JPEG2000 algorithms.

2011

Yan Michalevsky, Ronen Talmon and Israel Cohen, Speaker Identification Using Diffusion Maps, 19th European Signal Processing Conference (EUSIPCO 2011), Barcelona, Spain.

abstract

In this paper we propose a data-driven approach for speaker identification without assuming any particular speaker model. The goal in speaker identification task is to determine which one of a group of known speakers best matches a given voice sample. Here we focus on text-independent speaker identification, i.e. no assumption is made regarding the spoken text. Our approach is based on a recently developed manifold learning technique, named diffusion maps. Diffusion maps enable embedding of the recording into a new space, which is likely to capture the speech intrinsic structure. The algorithm is tested and compared to common dentification algorithms. Experimental results show that the proposed algorithm obtains improved results when few labeled samples are available.

Reuven Berkun, Ezri Sonn and Dmitry Rudoy, Detection of Score Changes in Sport Videos Using Textual Overlays, 7th International Symposium on Image and Signal Processing and Analysis (ISPA 2011), Dubrovnik, Croatia.

abstractproject

Sport videos are very popular content among consumers. Many of such videos include overlaid textual data about the game, such as score or player names. But, when the video is viewed on a small screen textual information becomes illegible. In this work we propose a fully automatic system for detecting changes in overlaid textual information, such as the game score. Our system detects text in the video and then marks the spatial and temporal coordinates of where the changes that have occurred in the text. This enables us to make text legible on a small screen by displaying an enlarged score box at the time of score change events. The proposed method was tested on a dataset of different sport videos, collected from YouTube. It achieves good results in terms of accurate spatial score box localization and accurate temporal score-change detection.

2010

Berdugo Guy, Soceanu Omri , Rudoy Dmitry , Dvir Itsik and Yair Moshe, Object Reidentification in Real World Scenarios across Multiple Non-overlapping Cameras, 18th European Signal Processing Conference (EUSIPCO 2010), Aalborg, Denmark.

abstractproject

In a world where surveillance cameras are at every street corner, there is a growing need for synergy among cameras as well as the automation of the data analysis process. This paper deals with the problem of reidentification of objects in a set of multiple cameras inputs without any prior knowledge of the cameras distribution or coverage. The proposed approach is robust to change of scale, lighting conditions, noise and viewpoints among cameras, as well as object rotation and unpredictable trajectories. Both novel and traditional features are extracted from the object. Light and noise invariance is achieved using textural features such as oriented gradients, color ratio and color saliency. A probabilistic framework is used incorporating the different features into a human probabilistic model. Experimental results show that textural features improve the reidentification rate and the robustness of the recognition process compared with other state-of-the-art algorithms.

Tamar Shoham and Yan Michalevsky, Fast H.264 Picture in Picture (PIP) Transcoder with B-slices and Direct Mode Support, 15th IEEE Mediterranean Electrotechnical Conference (MELECON 2010), Valletta, Malta.

abstractproject

H.264, the modern standard for video coding, has become increasingly popular in recent years, and offers solutions for many applications requiring video compression. Some of these applications require insertion of content into an already compressed video. This operation incurs high computational cost if a naive approach of full re-encoding is taken. Previous efforts in this field targeted the Baseline Profile of the H.264 standard. The Main Profile, that offers very efficient encoding modes using B-slices and Spatial Direct encoding, has not been addressed. We present Guided Encoding – a novel solution for efficient Picture-in-Picture embedding using the coding parameters of the original compressed bit stream. We also present an algorithm for handling Spatial Direct mode encoding to prevent propagation of errors caused by the embedding of new content into the original sequence. We reduce the computational time by an average factor of five in comparison to performing full re-encoding while preserving the video quality.

Omry Soceanu, Guy Berdugo, Dmitry Rudoy, Itsik Dvir and Yair Moshe, Where’s Waldo? Human Figure Segmentation Using Saliency Maps, 4th International Symposium on Communications, Control and Signal Processing (ISCCSP 2010), Limassol, Cyprus.

abstractproject

Human figure segmentation (HFS) is at the very core of many image and video processing tasks. Many solutions have been proposed for the separation of objects, or more specifically human figures, from image background in a video scene. Unfortunately, these solutions do not provide tight human segmentation in the most general conditions, so only a coarse segmentation of human figures can be assumed. In this paper it is assumed that a rectangular, not necessarily tight, segmentation of a human figure is available and a way to refine this segmentation is proposed. We use saliency detection for HFS in a single image, thus creating a mask that eliminates most of the background in the rectangular input while maintaining the human figure mostly intact. The proposed technique is a generalization over recently proposed saliency detection methods in order to better accommodate the special conditions of this specific problem. Tests show that saliency detection is beneficial for HFS. In addition, the proposed saliency detection is shown to improve HFS results and presents a viable solution for HFS in a single image or a video stream.

2008

Avihay Barazany, Royi Levy and Yekutiel Avargel, Lens-Motor Noise Suppression For Digital Cameras, 25th IEEE Convention of Electrical and Electronics Engineers in Israel 2008, Eilat, Israel.

abstractproject

In this paper, we formulate a speech enhancement problem under multiple hypotheses, assuming an indicator for the lens motor noise presence is available in the time domain. This approach is largely based on an existing speech enhancement algorithm, and was adapted to specifically handle lens motor noises. Hypothetical presence of speech together with an estimation of the stationary noise and the lens motor noise, are considered in the calculation of the desired signal spectral coefficients. Cost parameters control the trade-off between speech distortion and residual transient noise. An optimal estimator, which minimizes the mean-square error of the log-spectral amplitude, is derived, while taking into account the probability of erroneous detection. Experimental results demonstrate a substantial suppression of the motor noise with a minimal degrade to the perceived quality of the desired signal.

Guy Leibovitz, Assaf Elron, Zvi Ben-Haim and Yonina C. Eldar, Recursive Blind Minimax Estimation, 25th IEEE Convention of Electrical and Electronics Engineers in Israel 2008, Eilat, Israel.

abstractproject

We consider the problem of on-line (or recursive) parameter estimation in which, at each moment, an unknown deterministic parameter vector must be re-estimated from measurements corrupted by additive noise. We present efficient algorithms for calculating two variants of the blind minimax estimator, which is a biased estimator proven to outperform least squares in terms of mean squared error. These operate in the same setting as the recursive least squares (RLS) method and utilize it. Both algorithms have a computational complexity in par with RLS. We discuss the advantages and shortcomings of the presented methods and demonstrate through simulations situations in which they produce substantial gain over RLS.

Marlene Shehadeh and Oleg Kuybeda, Robust Nonuniformity Correction In Infrared Images Based On Global Motion, 25th IEEE Convention of Electrical and Electronics Engineers in Israel 2008, Eilat, Israel.

abstractproject

The presence of spatial and temporal fixed pattern noise (FPN) due to nonuniformity in sensor parameters is a well known problem in IR focal plane detector arrays. In order to obtain a noise-free image, the nonuniformity parameters should be estimated and compensated for. This work presents a scene-based nonuniformity correction (NUC) algorithm that estimates and compensates the nonuniformity parameters by equalizing output of different detectors that were exposed to the same information. The correspondence between pixels is found via an improved robust global motion estimation technique that locates and excludes local motion and occlusion regions from the estimation of motion parameters. The performance of the proposed algorithm is evaluated both in image sequences with simulated fixed-pattern noise and real infrared image sequences. Significant reduction in FPN is demonstrated in both cases.

Roee Sulimarski, Gal Gur-Arye, Avishai Adler and Yaakov Navon, Detecting Highlights And Notes On Printed Text, IEEE 25th Convention of Electrical and Electronics Engineers in Israel 2008, Eilat, Israel

abstractproject

The extensive use of digital cameras in creating an electronic database of document images, leads to a major problem in managing the vast amount of information. Many images include user-added marks, such as highlights and handwritten notes, which can be used in Content Based Image Retrieval. This paper presents a novel method for determining the existence of user-added marks in a given image and classification of existing marks. Two different approaches are integrated for the detection of marks: one based on color-spaces and the other on anomaly detection in text-dominant images. The proposed method achieved very successful results.

Dor Shabtay, Nissan Raviv and Yair Moshe, Video Packet Loss Concealment Detection Based on Image Content, 16th European Signal Processing Conference (EUSIPCO 2008), Lausanne, Switzerland.

abstractproject

Compressed video transmitted over unreliable channels such as wireless networks or the Internet may suffer from packet loss. A packet loss leads to image impairment that may cause significant degradation in image quality. In most practical systems, packet loss is detected at the transport layer and a decoder error concealment post-processor tries to mitigate the effect of lost packets. This helps to improve image quality but could still leave some noticeable impairments in the video. In some applications such as no-reference video quality evaluation, detection of concealment impairments is desired. If only video coding layer information is available, concealment has to be detected based on image content. In this paper, a novel technique for detecting impairments in a video sequence caused by packet loss after its concealment is proposed. The detection is based on image content only. Simulation results show good temporal and spatial concealment detection accuracy.

Raja Giryes, Alexander Bronstein, Michael Bronstein and Yair Moshe, Embedded System for 3D Shape Reconstruction, 3rd European DSP Education and Research Symposium (EDERS 2008), Tel Aviv, Israel.

abstractproject

Many applications that use three-dimensional scanning require a low cost, accurate and fast solution. This paper presents a fixed-point implementation of a real time active stereo three-dimensional acquisition system on a Texas Instruments DM6446 EVM board which meets these requirements. A time-multiplexed structured light reconstruction technique is described and a fixed point algorithm for its implementation is proposed. This technique uses a standard camera and a standard projector. The fixed point reconstruction algorithm runs on the DSP core while the ARM controls the DSP and is responsible for communication with the camera and projector. The ARM uses the projector to project coded light and the camera to capture a series of images. The captured data is sent to the DSP. The DSP, in turn, performs the 3D reconstruction and returns the results to the ARM for storing. The inter-core communication is performed using the xDM interface and VISA API. Performance evaluation of a fully working prototype proves the feasibility of a fixed-point embedded implementation of a real time three-dimensional scanner, and the suitability of the DM6446 chip for such a system.

Elad Roichman, Yaniv Solomon and Yair Moshe, Real-Time Pedestrian Detection and Tracking, 3rd European DSP Education and Research Symposium (EDERS 2008), Tel Aviv, Israel.

abstractproject

This paper presents an embedded prototype system for pedestrian detection and tracking in outdoor scenes. The system uniquely combines two common techniques for motion detection – temporal differencing and background subtraction. The results of these two techniques are adaptively averaged and thresholded to result in a binary motion image. Motion regions are efficiently segmented and non-target regions are removed. Detected pedestrians are tracked based on their location in previous frames. Both motion detection and tracking were designed with an embedded environment in mind and were optimized for the C6000 family of DSPs. A fixed-point version of the human detection and tracking algorithm was implemented using the DM642 EVM board. The proposed system and algorithm were verified to be accurate and robust by a set of experiments at real scenarios. A real-time implementation for D1 resolution shows the efficiency of the algorithm and proves the suitability of the DM642 chip for such a system.

2007

Yair Kuszpet, Dimitry Kletsel, Avi Levy and Yair Moshe, Post-Processing for Flicker Reduction in H.264/AVC, 16th Picture Coding Symposium (PCS 2007), Lisbon, Portugal.

abstractproject

The H.264/AVC standard mitigates some of the most noticeable artifacts of former video coding standards, such as blocking and ringing. However, it exhibits a new temporal artifact, not prevalent in these standards: a noticeable discontinuity between an intra frame and its preceding inter frame, known as flicker. This paper proposes a post-processing scheme for dealing with the flicker artifact. A motion compensated version of the intra frame is generated based on its preceding inter frame, and is used to filter the intra frame in order to reduce the discontinuity. In addition, a non-reference flicker measure is proposed, and is used as a basis for an adaptive flicker-reduction technique. Objective results, as well as subjective impressions, indicate that the proposed method significantly reduces the discontinuity, with an almost negligible drop in PSNR.

2005

Yair Moshe and Nimrod Peleg, Implementations of H.264/AVC Baseline Decoder on Different Digital Signal Processors, 47th International Symposium on Multimedia Systems and Applications (ELMAR 2005), Zadar, Croatia.

abstractproject

H.264/AVC is the newest video coding standard developed by the joint effort of ITU-T VCEG and ISO/IEC MPEG. This standard achieves a significant improvement in coding efficiency relative to former standards at the cost of increased complexity, thus gaining a lot of attention by industry, but creating a big challenge for efficient hardware and software implementations. In this paper three implementations of an H.264/AVC baseline decoder using different leading high-end Digitals Signal Processors are described. The implementation has been done, using the JVT reference software, by undergraduate seniors under the supervision of the authors, in the course of one year. This study is beneficial in several aspects. First, the complexity of an H.264/AVC decoder is measured and analyzed using “real world” hardware and software. Second, the difficulties that programmers could face when implementing an H.264/AVC decoder scheme are examined. And lastly, different Digital Signal Processors are examined under a very demanding algorithm.

2004

Daniel Kotlar, Zohar Idelson and Yizhar Lavner, Gene Prediction Analysis Based on Spectral Representation of DNA Sequences and Fuzzy Clustering, International Conference on Intelligent Systems for Molecular Biology (ISMB/ECCB 2004), Glasgow, United Kingdom.

abstractproject

We propose new measures for gene prediction in eukaryotes. The DFT at a frequency of 1/3 is computed for the four binary sequences for A, T, C, and G. The different distributions of the spectral phases in coding vs. non-coding regions are used to construct the Spectral Rotation measure for gene finding. Another proposed measure is based on representing each frame as a point in a three-dimensional complex space. The points corresponding to coding vs. noncoding regions tend to cluster differently. These discriminating features for gene prediction are shown to be potential candidates for locating short genes and exons.

Asaf Kagan, Yaakov Lumer and Oleg Kuybeda, Determining Hyperspectral Data-Intrinsic Dimensionality Via A Modified Gram-Schmidt Process, 23rd IEEE Convention of Electrical and Electronics Engineers in Israel, 2004, Tel Aviv, Israel.

project

Yair Moshe, Nimrod Peleg and Nadav Cohen, SPDemo – A Novel Software Tool for Teaching Multimedia Signal Processing, 2nd International Conference on Information Technology: Research and Education (ITRE 2004), London, UK.

abstract

The move from analog to digital media in recent years has given rise to the flourishing of digital signal processing techniques. In addition to the many benefits of this flourishing, the plenitude and sophistication of these techniques has caused learning and teaching the essence of this subject to be a more difficult task. In this paper a software tool for teaching multimedia signal processing is presented. This software tool is intended for assisting teaching and demonstrating speech and audio signal processing techniques. Using this tool one can select from a variety of techniques and parameters in a straightforward and simple way. The results of this processing can be played and viewed, numerically and graphically, in order to ease the analysis of the data. The purpose of this tool is to enable easy experimentation and demonstration of different algorithms and thus improve the quality of teaching.

2003

Oleg Kuybeda, Meir Bar-Zohar and Meir Shaked, Multiple Reference Frames Registration for Super Resolution, International Symposium on Optical Science and Technology 2003, Seattle WA, USA.

2002

Gidon Porat and Yizhar Lavner, Voice Morphing by 3-D Waveform Interpolation Surface and Lossless Tube Area Function, 22nd IEEE Convention of Electrical and Electronics Engineers in Israel 2002, Tel-Aviv, Israel.

abstractproject

Voice morphing is the process of gradually transforming the voice of a given speaker to that of another. The ability to change the speaker’s individual characteristics and produce high quality voices can be used in many applications. For example, in multimedia and video entertainment, voice morphing is just like its visual counterpart: while seeing a face gradually changing from one person’s to another’s, we can simultaneously hear the voice changing as well. Another application could be in forensic voice identification: creating a voice-bank of different pitches, rates, and timbres, to assist in recognition of the suspect’s voice. In this study we present a new technique, which enables the production of N intermediate voices that gradually change between voices of two speakers, or one voice signal that changes gradually. This technique is based on two components. One is creating a 3-D prototype waveform interpolation (PWI) surface from the residual error signal, which is estimated from LPC analysis, to produce a new intermediate excitation signal. The second component is a representation of the vocal tract by a lossless tube area function, and interpolation of the two speakers’ parameters.

2001

Yuval Cassuto, Michael Lustig and Shay Mizrachy, Real-Time Digital Watermarking System for Audio Signals Using Perceptual Masking, TI DSP & Analog Challenge, Dallas Texas, 2001., 1'st prize winner in TI DSP and Analog University Challenge

abstractproject

Recent development in the field of digital media raises the issue of copyright protection. Digital watermarking offers a solution to copyright violation problems. The watermark is a signature, embedded within the data of the original signal, which in addition to being inaudible to the human ear, should also be statistically undetectable, and resistant to any attempts to remove it. In addition, the watermark should be able to resolve multiple ownership claims (known as the deadlock problem), which is achieved by using the original signal (i.e., the unsigned signal) in the signature detection process.

In order to meet the above demands, a frequency-masking scheme using a psycho-acoustic model is used to ensure a maximal, yet inaudible, additive signature. The algorithm description is as follows: The audio signal is divided into segments. For each segment a local key is calculated and summed-up with an owners key (independent of the segment) to initiate a pseudo-random noise sequence for the segment. The noise is colored by a filter, which is calculated according to the psycho-acoustic model. The original signal is watermarked by adding the colored noise to it.

The project described hereafter contains MatlabTM simulation, a GUI C++ application (insertion & detection) and a real-time embedding scheme on the Texas Instruments TMS320C5410 DSP, using Spectrum Digital XDS510 EVM. The host-target communication was based on the JTAG connection, which caused a bottleneck for data transfer. In order to match real-time data transfer requirements the TIGER 5410/PC supporting Host-Port-Interface (using PC-ISA bus) is used.

Yizhar Lavner , Gil Bloch, Ophir Azulai , Irit Gazit ,Allen Goldblatt and Joseph Terkel, Detection and Discrimination of Sniffing and Panting Sounds of Dogs, IEEE International Conference on Acoustics Speech and Signal Processing (ICASSP 2001), Salt Lake City UT, USA.

abstractproject

Dogsí acute sense of smell enables them to accurately identify many types of odors. This ability serves in the detection of drugs and explosives, finding people, and other special tasks. The possibility of controlling dogsí behavior suggests many potential applications.

For this purpose, an automatic system for recognition of panting and sniffing according to their corresponding acoustic signals was developed. The system consists of three main components: analysis, training and discrimination. In the analysis stage the events were detected using short-time energy and zero-crossing rate functions. Features for discrimination were extracted utilizing an autoregressive model or by means of a Melfrequency cepstrum. A few parameters of the spectral envelope were selected, based on their discrimination ability. In the training stage, a feedforward neural network was trained, using labeled signals. Alternatively, a fuzzy K-nearest neighbor algorithm can be used. Correct identification of up to 96.8% was achieved in the discrimination stage.

2000

Tal Mizrahi, Eran Borenstein, George Leifman, Yuval Cassuto, Michael Lustig, Shay Mizrachi and Nimrod Peleg, Real-Time Implementation for Digital Watermarking in Audio Signals Using Perceptual Masking, 3rd European DSP Education and Research Conference (ESIEE 2000), Paris, France.

abstractproject

The advancement of digital audio enables the reproduction of copyright-protected media without any degradation in audio quality. This causes severe problems of copyright violations and emphasizes the need for copyright protection mechanisms. Digital watermarking is one of the most popular copyright protection mechanisms. In this paper we present a novel real-time signature embedding system for embedding of digital watermark in audio signals. The signature detection is done using off-line Windows 9x/NT application (this application enables embedding and detection of digital signature). The embedding mechanism enables an owner to insert into its own media a digital watermark in such a way, that the audio quality is not reduced. This is done, by using the human auditory system’s masking characteristics. The detection mechanism enables the owner to check for the existence of the watermark in a tested media. The presented system solves the deadlock problem (multiple ownership claims) by keeping the original media or parts of it for future ownership claims.

Nira Shezaf, Hagit Abramov-Segal, Ilan Sutskover and Ran Bar-Sella, Adaptive Low Complexity Algorithm for Image Zooming at Fractional Scaling Ratio, 21st IEEE Convention of the Electrical and Electronic Engineers in Israel 2000, Tel Aviv, Israel.

abstractproject

Interpolation of images to higher resolution is an important feature in many digital imaging applications ranging from image zooming for digital camera to image zooming in digital video. Practical digital imaging applications require the ability to scale an image in any fractional ratio, under low complexity/memory restrictions. Many traditional zooming algorithms are restricted to integer zooming ratios and yield poor results in edgy areas of the image; some are even restricted to powers-of-two ratios. Algorithms that achieve high quality are usually too complex to be implemented in low complexity/memory environments like VLSI. In this work we present a new adaptive algorithm for image zooming, suitable for fractional zooming ratios, motivated by the work of Ting. et a1.[3]. The new algorithm performs better than known algorithms on a variety of images at both integer and fractional zooming ratios and is better suited for VLSI implementation.

Ofir Shalev, Leon Schveiger, Ilan Sutskover and Ran Bar-Sella, A Modified JPEG-LS Image Coder With Increased Channel Error Robustness, 21st IEEE Convention of the Electrical and Electronic Engineers in Israel 2000, Tel Aviv, Israel.

abstractproject

JPEG-LS is a lossless and near-lossless compression algorithm for continuous-tone images. As many variable length source coders, it is highly vulnerable to channel errors. In this paper, some modifications to the JPEG-LS are suggested. The modified algorithm has good robustness to channel errors, while maintaining compression ratio that is close to that of the original JPEG-LS.

1998

Hanan Leizerovich, Eyal Feigenbaum and Hagai Krupnik, Development of an Ancillary Instrument for Piano Tuners, 9th Mediterranean Electrotechnical Conference (MELECON 1998), Tel Aviv, Israel., Participated in IEEE R8 final student papers contest

project

1996

Ran Bar-Sella, Nir Shachar, Zelig Weiner and Nimrod Peleg, MPEG-1/VCD Digital Video Editing, 7th International Conference on Signal Processing Applications and Technology (ICSPAT 1996), Boston MA, USA.

project

1993

Raz Bareket, Yoram Or-Chen, Nimrod Peleg, Guillermo Sapiro and Omer Tzifroni, Fractal Image Compression Using a Motorola DSP96002, 4th International Conference on Signal Processing Applications and Technology (ICSPAT 1993), Santa Clara CA, USA.
Project

project