IWSSIP 2001></td>
    </tr>
    <tr>
      <td width=

June 7 - 9, 2001 - Bucharest, Romania



Call for PapersPaper SubmissionWorkshop Deadlines
Workshop TopicsRegistrationProgram
Tutorials CallTutorialsSocial Program
Workshop OfficialsOrganisersCo-Sponsors
Workshop Site

Paper abstracts

Plenary Session P1  

          INVARIANT FEATURE EXTRACTION BASED ON THE HOUGH TRANSFORM

J. Turán, K. Fazekas*, P. Farkaš, D. Šiškovičová

Department of Electronics and Multimedia Telecommunications University of Technology Košice Park Komenského 13 041 20 Košice, Slovakia

* Department of Microwave Telecommunications University of technology Budapest Goldmann Tér 3 1111 Budapest, Hungary

e-mail: jturan@tuke.sk, kfazekas@cyberspace.bme.hu

Abstract: In this paper a system for invariant object recognition based on the Hough transform, correlation, N transform and the neural network classificator is proposed. The system was implemented as a programme package on PC and tested in some recognition experiments with original and noisy images.

GENETIC SIGNALS: AN EMERGING CONCEPT

Paul Cristea
"Politehnica" University of Bucharest, RO

Abstract: The almost complete sequencing of the human genome, as well as the public access to most of its content, offer tremendous opportunities to explore in depth its content and to data mine this unique information depository. The classic approach of representing DNA as symbolic sequences of nitrogenous bases or of symbolic codons encoding polypeptide chains essentially limits the methodology of handling the information to mainly pattern matching procedures. Converting the DNA sequences into digital signals using a base 4 representation of the nitrogenous bases leads to the conversion of the codons into numbers of the range 0-63 and of the amino acids, together with the terminator, into numbers of the range 0-20. Correspondingly, this leads to the conversion of DNA sequences into digital genetic signals and opens the possibility to apply a whole range of powerful signal processing methods for their analysis.

Currently, only about 32000 genes containing the instructions to make proteins, but representing only about 5 percent of the human genome, are considered of interest. The vast majority of the genome is considered junk, as it has been discovered that it contains a large amount of mobile (transposable) elements that bear a close resemblance to the DNA of independent entities like viruses and bacteria. Similarly to the mitochondria, that have also started their ancestral life as independent entities, to become the main energy suppliers of the eukariotic cells, a significant part of the extra-genic chromosomal DNA has very probable an important role in the control of protein synthesis. The paper uses the ICA approach to search for independent components in the extra-genic DNA. A special attention is given to fundament the choice of a "natural" correspondence between the nitrogenous bases and the digits in base four (Thymine = 0, Cytosine = 1,Adenine = 2, Guanine = 3).

Independent Component Analysis (ICA) is a special case of the Blind Separation of the Sources (BSS) method. Its goal is to recover statistically independent source signals from some available linearly mixed signals produced by an unknown medium. Many applications are actively being developed, including speech recognition, telecommunications, bio-medical signal and image processing. For complex real-life problems, the computational load becomes excessive, especially when unknown relative shifts of the independent components have to be considered. The paper present some preliminary results of a parallel implementation of the ICA problem, suitable for the cases in which the dynamics of the system can not be ignored.

Special Session S1

WAVELETS AND STANDARDS: AN OVERVIEW

P. Schelkens and J. Cornelis

Vrije Universiteit Brussel/IMEC Dept. ETRO Pleinlaan 2 - B-1050 Brussel - Belgium {pschelke,jpcornel}@etro.vub.ac.be

Abstract: Wavelet-based compression algorithms have recently become extremely popular. In this paper two ISO/IEC international standards utilizing this technology are reviewed, i.e. JPEG2000 and MPEG4. Additionally, the prospects for further adoption of wavelet technology in multimedia standards are discussed.

EFFICIENT IMPLEMENTATIONS OF WAVELET TRANSFORMS -A ROADMAP

Y. Andreopoulos, P. Schelkens, T. Stouraitis*, J. Cornelis

Vrije Universiteit Brussel/IMEC Dept. ETRO Pleinlaan 2 - B-1050 Brussel – Belgium, {yandreop,pschelke,jpcornel}@etro.vub.ac.be

* University of Patras - Dept. ECE VLSI Design Laboratory Rio 26500 - Greece

Abstract: The two major implementation methods for the discrete, two-dimensional binary-tree wavelet decomposition are presented. They are proposed in the context of efficient coupling with coding algorithms of compression standards, namely JPEG-2000 and MPEG-4. When implemented in software or hardware systems, they are capable of producing in real-time the binary-tree decomposition of the entire input image with a higher sample-rate. This is achieved by dividing and localizing the processing into small blocks of data. These blocks can efficiently be handled by a cache hierarchy in a programmable processor or by a custom-hardware design.

MPEG-4 VISUAL TEXTURE CODING: VARIFORM, YET TEMPERATELY COMPLEX

Gauthier Lafruit, Bart Vanhoof

IMEC/DESICS, Kapeldreef 75, B-3001 Heverlee Belgium

Abstract: Modelling the arithmetic complexity of state-of-the-art algorithms enables predicting their per-formance for a specific implementation. Such performance analyses have - apart from the algorithmic functionality - be taken into account when selecting algorithms in standards such as MPEG-4. This paper presents an arithmetic complexity analysis of the scalable Visual Texture Coding tool within MPEG-4.

IMPLEMENTATION OF A SCALABLE WAVELET MPEG-4 COMPRESSION SYSTEM

Adrian Chirila-Rus, Bart Vanhoof and Henk Corporaal

Inter-university Micro Electronic Centre (IMEC) Kapeldreef 75, B-3001 Leuven, Belgium, E-mail; chirila@imec.be

Abstract: In the MPEG-4 standardisation process the wavelet based compression scheme has been adopted for still images and texture coding. Special attention is devoted to the quality of services (QoS) and the feature of graceful degradation, i.e. the scalability of the media content according to the client’s bandwidth and processing power. This paper presents as scalable wavelet compression system on a chip. Our hardware implementation of the Ozone chip and of its next generation chip FlexWave, proves the functionality, performances and utility of a scalable compression system dedicated for still images and texture coding.

Plenary Session P2

IMPLICATION OF BROADBAND RADIO ACCESS TO UMTS

B. Zovko-Cihlar, M. Suknai * , M. Grgic

Faculty of Electrical Engineering and Computing, University of Zagreb, Unska 3 / XII, HR-10000 Zagreb, CROATIA

* Croatian Telecom, Mobile Networks Technology and Services Development Department, Draskoviceva 26, HR-10000 Zagreb, CROATIA

E-mail: martina.suknaic@ht.hr; branka.zovko@fer.hr

Abstract: The Universal Mobile Telecommunication System (UMTS) - third mobile communication standard developed by ETSI was planned to provide bit rates up to 2 Mbit/s for indoor coverage (till 10 meters, for office and home environments) for low mobility user and up to 384 kbit/s for outdoor coverage (till 20 km, for suburban and rural areas) for high mobility users. The existing and future Internet and telecommunication services have demands for higher bit rates and wider bands than UMTS as 3 rd generation technology can provide. There are new projects in ETSI and around the world, which research and develop standards for fixed and cordless broadband radio accesses.

    APPLICATION OF GABOR FILTERS FOR EFFICIENT IMAGE REPRESENTATION

M. Grgic, S. Grgic, M. Ghanbari *

Faculty of Electrical Engineering and Computing, University of Zagreb, Unska 3 / XII, HR-10000 Zagreb, CROATIA

* Department of Electronic Systems Engineering, University of Essex, Colchester, Essex, CO4 3SQ, UNITED KINGDOM

E-mail: mgrgic@ieee.org

Abstract: In this paper we show the application of Gabor wavelet in image features extraction. Using Gabor wavelets an image is decomposed into M ×L filtered images. For each filtered image the mean value and standard deviation of the energy distribution of the Gabor transform coefficients are computed. An image is described by a feature vector with M ×L feature parameters. Gabor feature extraction is applied on a large image database and retrieval results are presented. Our results show that application of Gabor filters in image retrieval system provide high retrieval accuracy.

Session S2

SEGMENTATION OF THE LEFT VENTRICLE OF THE HEART IN VOLUMETRIC CARDIAC MR IMAGES: EVALUATION STUDY OF CLASSICAL AND GVF-SNAKES.

A. I. Gavrilescu, R. Deklerck, A. Salomie, P. Dendale*, J.Cornelis

Vrije Universiteit Brussel, Dept ETRO-IRIS Pleinlaan 2; 1050 Brussels; Belgium

*University Hospital Antwerp Dept. Cardiology Wilrijkstraat 10, 2650 Edegem; Belgium

e-mail: aigavril@etro.vub.ac.be

Abstract: The deformable models, known as "snakes", have a wide applicability in image segmentation. However it is an approach that cannot be applied without a prior tuning phase. Depending on the kind of images the user wants to analyse, he will have to (1) choose a pre-processing method to eliminate for instance the noise in the image and enhance the edges, (2) select an edge detector to obtain a suitable force field map and (3) tune the elasticity/rigidity parameters of the snake-model to obtain boundaries, which follow closely the real boundaries in the image but not the jags in the edges that are due to artefacts or noise. In this paper we discuss the results obtained with the snake-based approach for a particular segmentation task: i.e. the segmentation of the left ventricle of the heart from a time-series of volumetric cardiac MR images. We show that pre-processing these images with an anisotropic diffusion algorithm, considerably improves the quality of the segmentation. Further we compare the results obtained with a classical snake method, deriving the force-field as the gradient of the modulus of the vertical and horizontal Sobel edge detector and an enhanced snake-based algorithm (Gradient Vector Flow), computing the force field from the binary image produced by the Canny edge detector. The best results are obtained for anisotropic diffusion combined with a Canny edge detector, and with the GVF-snake algorithm.

AUTOMATIC HYBRID SEGMENTATION SCHEME FOR THE EXTRACTION OF VISUAL DOMINANT COMPONENTS

Constantin Vertan, Vasile Buzuloiu, and Nozha Boujemaa*

Image Processing and Analysis Laboratory, Bucharest Politehnica University, ROMANIA

(*) IMEDIA Group, INRIA Rocquencourt, FRANCE

Abstract: An inherent problem of most image segmentation algorithms is the requirement of an a priori knowledge regarding the number of object types (segmentation classes) within the scene. The proposed segmentation algorithm involves both partitional and hierarchical clustering approaches, in order to compensate some drawbacks of each individual method; the final segmentation result is an approximate image partitioning in visual important regions. The hybrid scheme embeds an automatic procedure for the stopping of the hierarchical class number reduction. The stopping criterion is adaptively computed according to the image contents descriptors, in order to effectively reject any still existing small, unimportant classes. The algorithm is fully automatic, no user interaction being needed.

 

SEGMENTATION OF DUAL-BAND IMAGES OF X-RAY CHICKEN BREAST USING A COMPETITIVE HOPFIELD NEURAL NETWORK

Catalin Amza

De Montfort University, Engineering and Technology Department,

The Gateway, LE1 9BH, Leicester, UK

Fax: +44 116 257 7099; acata@dmu.ac.uk

Abstract: The contaminant detection process of a food product is an important stage of a modern food production factory. The large demand of bone-free for meat has lead producers to use automated systems. One such system is the automated detection of bones from raw chicken breast meat. An X-ray image of the breast meat product is taken and analysed by the system. The most important step of the process of inspecting that product is the segmentation of the image into meaningful objects (bones). This paper presents a Competitive Hopfield Neural Network for the segmentation of dual-band X-ray images of raw chicken breast meat. A back-propagation neural network or a simple look-up table can then be used to assign the segments into desired categories (bones, or non-bone objects).

Session S3

MONITOR CALIBRATION AND COLOUR MANAGEMENT

L. Mandic , S. Grgic*, M. Grgic*

Faculty of Graphic Arts, Getaldiceva 2, HR-10000 Zagreb, Croatia

* Faculty of Electrical Engineering and Computing, Unska 3 / XII, HR-10000 Zagreb,

E-mail: mandic@grf.hr; sonja.grgic@fer.hr

Abstract: The aim of this paper is to show the importance of monitor calibration and colour management. While the calibration of the monitor was a serious task only for desktop publishing (DTP) users, today the requirements becomes larger due to internet homepages. The calibration of monitor doesn't effect only on image reproduction, even on the readability of text on Web sites. However we come to one point, where only calibration methods, as the fundamental part of every monitor setup, are not sufficient any longer, and a colour management system with its profiling and matching features is required.

 

PERFORMANCE MEASURE OF OPTICAL COMMUNICATION SYSTEM BASED ON LASERS IN IMAGE TRANSMISSION

I. Kuzmanić, I. Vujović

Split College of Maritime Studies, Zrinsko-Frankopanska 38, 21000 Split, Croatia

Abstract: Measures of optical system performance are considered in this paper. The MatLab code for just-noticeable-distortion is given. Some aspects of laser systems safety are discussed. Example applications of optical systems in LAN is given.

REACTIVE CONGESTION CONTROL FOR MULTIPOINT VIDEO SERVICES

R. Dobrescu and M. Dobrescu

POLITEHNICA University of Bucharest, Splaiul Independentei 313, Romania

E-mail: radud@aii.pub.ro

Abstract. The paper proposes a new service architecture for real-time multicast video. To support this service a multipoint feedback mechanism is introduced and studied. This mechanism solves to problems specific for multipoint communications: the problem of the amount of feedback reduction, due to the increase of the number of destinations from the same source description, solved by using a feedback polling mechanism and the problem of the available bandwidth variation, because various branches of a multipoint connection have different amounts of available bandwidth, solved by using a combination between the adaptive encoding and the feedback-based rate control. For the successful use of the first mechanism, two algorithms are developed: one to locate the most congested branch in the multicast tree, the other to observe the overall congestion state of this tree. In order to meet the target high priority cell rate a special encoding technique, based on an adjustment of the encoder’s quantization parameter is described. This procedure achieves the control of the total output rate from the video encoder. In addition, data partitioning is utilized to control the output rate of low and high priority information. The performances of the proposed feedback mechanism, associated with the encoding algorithm, were investigated on a series of simulations on a point-to-multipoint network model with seven ATM switches interconnected in a binary tree topology. The simulation results showed the scalability with the number of destinations of the cell rate and of the video signal quality and the impact of the propagation delay and of the number of congested path, proving that video feedback mechanism is capable of providing better quality than a classical CBR service at the same amount of available bandwidth and can adapt to varying degrees of congestion in the network in order to improve video quality.

Session S4

A NEW PARALLEL ARCHITECTURE TO DEAL WITH REAL-TIME COMPUTER IMAGE PROCESSING

V. Ila, P. Ridao, j. Batlle

Institute of Informatics and Applications. University of Girona

Avda. Lluis Santaló s/n. 17003-GIRONA, SPAIN

tel: 34 972 418767 fax: 34 972 418098

email: (viorela,pere, jbatlle @eia.udg.es)

Abstract: This paper will describe an architecture composed by specific cells real-time image processing oriented. The architecture has an optimal relation cost/efficiency, modularity, application based dynamic re-programmability as well as the facility of interconnection, extension, and synchronisation of the system, that makes possible the interchange of information between any pair of cells of the network. This architecture is design to be able to carry out parallel operations in computer vision processes , specially in the phases of pre-processing, processing and recognition.

ON THE INTRODUCTION OF A MULTIRESOLUTION FUZZY RUNLENGTH IMAGE DESCRIPTION SCHEME

Constantin Vertan, Vasile Buzuloiu, and Christine Fernandez-Maloigne*

Image Processing and Analysis Laboratory, Bucharest Politehnica University,

(*) IRCOM-SIC, University of Poitiers, FRANCE

ABSTRACT: Image retrieval systems are based on effective image description schemes that embed most of the visual cues characterizing the image: color, texture, structure. Successful descriptors take into account both the color balance and the spatial color distribution, spanning several spatial resolution levels. We propose to extend the classical runlength image description by embedding some fuziness degree in both the color description and the runlength description. Multiresolution is performed with respect to the allowed degree of inter-color confusion, equivalent to the usual spatial multiresolution approach. The performed tests proved that the proposed method overcomes the classical runlength approach and has a good potential for texture recognition and general purpose image retrieval.

 

THREE-DIMENSIONAL VECTOR LUM SMOOTHERS

R. Lukáč, and S. Marchevský*

(*) Department of Electronics and Multimedia Communications,

University of Technology Košice, Park Komenského 13,

041 20 Košice, The Slovak Republic

e-mail: lukacr@tuke.sk, marchs@tuke.sk

Abstract: The noise filtering of color image sequences is very important and interesting problem since high dimensionality of processed image signals. According to motion and color information, color image sequences represent three-dimensional vector-valued image signals and thus, for the processing of color image sequences, the three-dimensional vector filters provide optimal approach, only. On that reason, this paper is focused on the impulse noise suppression by a new class of vector LUM smoothers with a three-dimensional (3-D) filter window. To provide objective measures of the proposed method, four objective criteria for signal-details preservation, noise suppression, motion preservation and color distortion are used.

Session S5

RADAR TARGET TRACKING BY NEURAL KALMAN FILTER

Ján OCHODNICKÝ, Jozef TKÁČ, Štefan ŠPIRKO

Department of Radar Technology, Military Academy,

P.O.Box 76, 03101 LIPTOVSKÝ MIKULÁŠ, Slovak Republic

E-mail: ochodnicky@valm.sk

Abstract: Radar is designed to perform accurate target tracking in direction and range. The modern radar systems use digital signal processing. These new trends of signal and data processing are implemented in military radar of Slovak army too.

A target tracking is a fundamental building block of majority Radar Data Processing (RDP) algorithms. The target tracking is an important issue in military surveillance systems, especially when such systems employ multiple sensors to interpret the environment. Typical sensors are radar, infrared systems, electronic support measurements, sonar etc. The objective of target tracking or, in general, multiple-target tracking (MTT) is to partition sensor data into sets of observations and produce track of a target.

The most popular approach in MTT is the Kalman filtering that is effective for simple scenery such as a clutterless environment or a single sensor tracking a single target. The basic problem of Kalman filtering in nonlinear dynamics systems application is the regulation of the Kalman gain. The paper presents the concept of the proposed backpropagation neural network and Kalman filter to reduce the estimation error due to an imperfection of a radar system and tracked target.

The paper describes processing of real input data. The backpropagation neural network implemented in the output of the Kalman filter reduces a final error. Input real data for tracking was obtained from the modern radar named MORAD.

  RADAR OBJECT RECOGNITION BY WAVELET TRANSFORM

Ing. Jozef Tkáč, PhD.

Department of Radiolocation, Military Academy

P. O. Box 45/204, 031 19 Liptovský Mikuláš, Slovak Republic

e-mail: tkac@valm.sk

Abstract: This paper presents method of radar object recognition by wavelet transform and neural network. Analyzing range profiles by wavelet transform can show information about tracking object more clearly. It also gives a method of constructing feature vector for automatic recognition, which can make the dimension of feature vector be much smaller than dimension of primitive echo signal. A three-layer backpropagation neural network is used as recognizer. This problem is aimed to radar systems with the duration of transmitted pulses longer than size of object.

MORPHOLOGICAL ASSOCIATORS FOR PATTERN RECOGNITION

Yiannis Boutalis, Basil G. Mertzios, and Adrian G. V. Moise*

Democritus University of Thrace,

67100 Xanthi, Greece

E-mail: ybout@ee.duth.gr,

mertzios@uom.gr

*Oil and Gas University of Ploiesti,

Bd. Bucuresti, Nr. 39, 2000 Ploiesti, Romania

E-mail: amoise_98@yahoo.com

 

Abstract: By using two theorems on Morphological Associative Memories (MAM's), a well suitable to pattern recognition mathematical background is developed. The ability of human beings to retrieve information based on associated cues continues to elicit great interest among researchers. Investigations of how the brain is able to make such associations from partial information have lead to a variety of artificial neural networks models that act as associative memories. Thus far MAM's have been used in two different ways: a direct approach which is suitable for input patterns containing either dilative or erosive noise and an indirect one for arbitrarly corrupted input patterns which is based on kernel vectors. In this paper, we establish the proofs for all the claims made about the choice of kernel vectors and perfect recall in kernel applications. Moreover, we provide arguments for the success of both approaches beyond the experimental results presented up to this point.

Plenary Session P3

SIGNIFICANCE OF BACK-SCATTERED LIGHT INTENSITY FOR QUANTITATIVE OPTICAL MEASUREMENTS IN TISSUE

M. Kessler, M. Boehnert

Institute of Physiology and Cardiology, University of Erlangen-Nuremberg, Waldstr. 6, 91054 Erlangen, Germany

e-mail: kessler@ipk.med.uni-erlangen.de

 

Abstract: The strategy applicable for precise analysis of tissue function is the use of experimental investigations of capillary and cellular mechanisms in human and mammalian organs. By use of EMPHO-SSK Oxyscan it is possible to produce high quality 3D-images of functional structures of subcellular space All tissues are complex multi-component systems which usually cannot be solved primarily by application of mathematical tools. Spatial patterns of real functional structures can be reconstructed mathematically step by step based upon the data obtained by experiments. The finite solvation will be composed of a series of algorithms depending on the size and number of subsystems that must be applied for a solution of interest.

 FINANCIAL TIME-SERIES PREDICTION USING SECOND-ORDER PIPELINED RECURRENT NEURAL NETWORK

Abir Hussain and Panos Liatsis*

Department of Computer Science

De Montfort University, Leicester

Email: abirh@dmu.ac.uk

* Department of E&EE

UMIST, PO Box 88, Manchester

Email: panos@csc.umist.ac.uk

Abstract: In this paper, we propose a new type of higher-order pipelined recurrent neural networks called the second-order pipelined recurrent neural network. The purpose of the network is to improve the performance of the pipelined recurrent neural network by accommodating second order terms in the inputs. The proposed network was used in the prediction of financial time-series where an improvement in the signal to noise ration of 1.92 dB was achieved when compared to the pipelined recurrent neural network.

  Session S6

CLASSIFYING MYOELECTRIC SIGNALS BY USING CORRELATION METHOD

Ahmed O. Abdul Salam, Yousef Al-Assaf,

Electrical, Electronics, and Computer Engineering Department,

American University of Sharjah, P.O.Box: 26666, Sharjah, United Arab Emirates,

aoasalam@aus.ac.ae, yassaf@aus.ac.ae

 

Abstract: This paper investigates the classification of Myoelectric Signals (MES) using correlation Techniques. While the classification of the main control movements ot the forearm flexion and extension is considered in this work, the ultimate goal is to improve myoelectric system control performance. The statistical approach of matching via correlation factor is used for the classification and timing of the two movements. Results have shown that this simple correlation method resulted in adequate classification performance.

DIAFRAGMATIC EMG SIGNAL ANALYSIS AND SIMULATION

M. Ungureanu, W. Wolf* , R. Strungaru

Polytechnica University of Bucharest
SPL. Independentei 313, 77206 Bucharest, Romania

* Universität der Bundeswehr München
Werner-Heisenberg-Weg 39,
D-85579 Neubiberg, Germany, Werner.Wolf@unibw-muenchen.de

 

Abstract: The inspiratory diaphragmatic EMG (EMGdi) contains altering ECG signal. In order to determine the efficiency of the denoising methods simulated EMGdi data are needed. For this simulation an additive model is proposed, based on a real EMGdi signal. The study uses 5 real EMGdi and ECG data recorded with a sampling frequency of 2000 Hz for each channel.

  ELECTRONEUROGRAM SIGNAL DETECTION WITH THIRD ORDER STATISTICS

A. D. Ionita

Automation and Computer Science Faculty, "Politehnica" University of Bucharest,

Spl. Independentei 313, Bucharest, 77206, Romania

ancai@ciid.pub.ro

Abstract: The paper studies the detection of electroneurogram signals from measurements containing additive noise. A simulation was done in order to prove that averaging loses many responses to stimuli, if the signal-to-noise ratio is very low. Then, the median nerve electroneurogram bispectrum was computed and was used for reconstructing the clean ENG signal, without the Gaussian contribution that is inherently present in bioelectrical recordings. This technique was used for extracting the first individual responses from real data.

  Session S7

SYNCHRONIZATION OF CNN-BASED CHUA CIRCUITS FOR SECURE COMMUNICATIONS IN R. F. CHANNELS

F. Beritelli(a), G. Di Marco(b), L. Fortuna(b), M. Francese(c)

a DIIT
University of Catania

b DEES
University of Catania

Viale A. Doria 6 - 95125 Catania - Italy

c MTI
Via G. Leopardi 41-95127 Catania

Abstract: This paper presents a study of the synchronization of two chaotic circuits based on Cellular Neural networks (CNN) for secure communications on radio frequency (RF) channels. The synchronization method used is based on division into sub-systems. Of the various chaotic circuits present in literature (Duffing, Chua, Lorenz circuit, etc.) we chose the Chua circuit. In order to overcome the limits of the traditional Chua circuit (dynamic speed, bandwidth, etc.) we used a Chua circuit based on CNNs. A 433 MHz F.M. modulator/demodulator was used in the testing phase. The correct synchronization between two Chua circuits linked by a wireless channel was verified by a digital oscilloscope connected to a PC acquisition board.

A NEW WAY FOR NEURAL NETWORKS TRAINING

I. Nastac

Electronics and Telecommunications Department
"Politehnica" University of Bucharest

E-mail: nastac@danube.euroqual.pub.ro

Abstract: The main purpose of the present paper is to establish how a viable Artificial Neural Networks (ANN) structure at a previous moment of time could be re-trained in an efficient manner in order to support modifications of the initial project. In view of this, we use an anterior memory, scaled with a certain convenient value. The evaluation of the computing effort involved in the retraining of an ANN show us that a good choice for scaling factor could substantially reduce the number of training cycles irrespective of learning methods.

EVALUATION OF SRM MAGNETIZATION SURFACE USING NEURAL NETWORKS

V. Trifa, V.T. Dadârlat, A. Peculea

Technical University of Cluj-Napoca
15 C. Daicoviciu St., RO-3400 Cluj-Napoca, Romania
Adrian.Peculea@edr.utcluj.ro

Abstract: The switched reluctance motor (SRM) is known at present as one of the most performing driving motors of low and medium power. One of the basic problems that preoccupies the specialists is the establishing of the electromagnetic parameters of SRM, as a premise for developing methods of designing and modeling driving systems with this motor. The evaluation of the magnetization surface solves the problem of the SRM modeling by taking into account the saturation which, in the case of this motor, is number one non-linearity. The determination of electromagnetic parameters is based on the measurement of SRM magnetization curves. The method used is current decay. The description of magnetization surface as a mathematical expression is an extremely laborious task. That is why the best solution to solve this task is the use of an artificial neural network. In order to solve this problem, it has been defined a backpropagation - type artificial neural network with two inputs – current and angle – and one output – flux.

Session S8

A ROBUST AND SIMPLE ISOLATED WORD RECOGNITION ALGORITHM IMPLEMENTED ON "ACTIVE TEMPLATE LIBRARY" TECHNOLOGY

F. Beritelli, S. Serrano*

DIIT - University of Catania
V.le A. Doria, 6 – 95125 CATANIA (ITALY)

* MTI
Via G. Leopardi 41 – 95127 CATANIA (ITALY)

Abstract: This paper presents a binary component for isolated word recognition developed in the Microsoft Visual C++ 6.0 environment using ATL (Active Template Library) technology. ActiveX is extremely simple and takes up only 127 kbytes. Its performance is also very high as it implements new speech recognition algorithms based on techniques that exploit template-based pattern-matching and are very robust to background noise.

FUZZY VARIANTS FOR SPEECH RECOGNITION ALGORITHMS

Inge Gavat, Zica Valsan, G.Ovidiu, B. Sabac, M. Zira

Politechnica University of Bucharest, Splaiul Independentei 313, 77206, Bucuresti, Romania.

e-mail: gavat@alpha.imag.pub.ro

 

Abstract: In the paper there are presented fuzzy variants of well known classification algorithms for speech recognition based on speech characterization by melcepstral coefficients and the first and second order differences. Improvements from 1-3%in speech recognition rate versus non-fuzzy version are a proof of a better adaptation of such algorithms to the dificulty of the classification task.

  WAVELETS IN BIOMEDICAL SIGNAL PROCESSING

S. Stanković, D. Milovanović, U. Batričević, and R. Maksimović*

Faculty of Electrical Engineering , University of Belgrade, Yugoslavia

*School of Medicine, University of Belgrade, Yugoslavia

Abstract: Wavelet techniques are applied to biomedical signal processing, including denoising, enhancement, compression and segmentation of medical images. Two novel algorithms are proposed for image denoising and segmentation. Experimental results demonstrate high performance of the presented methods.

  Session S9

SINGLE-QUADRANT ANALYTIC IMAGES FOR 2-D DISCRETE WIGNER-VILLE DISTRIBUTION

R. Iordache A. Beghdadi*

Signal Processing Laboratory
Tampere University of Technology
FIN-33101 Tampere, Finland

email: iordache@cs.tut.fi

*L2TI-Institute Galil´ ee
Universit´ e Paris XIII
FR-93430 Villetaneuse, France

email: beghdadi@l2ti.univ-paris13.fr

Abstract: The cross-term interference and the aliasing prevent Wigner-Ville distribution (WVD) to be the ideal tool in spatial/spatial-frequency analysis of real images. A solution to reduce and even eliminate these artifacts is the use of 2-D analytic images in computing the 2-D WVD. The first objective of this contribution is to present the Wigner-Ville distribution from the point of view of image analysis, i.e. to emphasis the desired properties, to identify the introduced artifacts and at the same time to present solutions to minimize their effect. The second objective is to show that the use of single-quadrant analytic images eliminates the aliasing and highly reduces the cross-term interference, and can provide a spatial/spatial-frequency representation having the same spatial-frequency resolution and support as the analyzed real image.

  A NEWTON-RAPHSON ITERATION ALGORITHM FOR 3-D IMAGE RECONSTRUCTION

Martin Breznan , Martin C.B.Smith *

Telecommunication Department University of Zilina, Slovakia
breznan@fel.utc.sk

* Department of Electrical Engineering University of East London, United Kingdom
msmith@iee.org

Abstract: This paper describes an implementation of the Newton-Raphson iteration algorithm used for 3-D image reconstruction from binocular stereo images. The algorithm is based on the gradient of disparity. It improves results of 3-D reconstruction, which can be made by different methods, for example, by correlation. It can also increase the speed of computation. The results of experiments with real stereo pictures are presented.

  FRACTAL CHARACTERIZATION OF NON-GAUSSIAN CRITICAL MARKOV RANDOM FIELDS

R. Ghozi

Dept. Applied Math and Digital Communications, Ecole Superieure des Communications, Tunis, Tunisia. E-mail: raja.g@cynex.com

ABSTRACT: In this paper we propose a fractal characte rization of a class of non-Gaussian Markov Random Fields (MRFs). In particular, we show that all order statistics of this class of MRFs are power-law functions. We then present a comparative study between the class of CMRFs and that of fractional Brownian motions (FBMs). Both of these classes of models can be used to describe self-similar phenomena. Their characteristics, namely the index of similarity of a FBM, and the critical exponents of a CMRF, are defined and contrasted. We argue that CMRFs provide a more flexible mechanism for generating self-similar patterns, since the parameters of a CMRF can be selected to generate non-Gaussian anisotropic patterns, while FBM models are inherently isotropic and Gaussian. This research was partially supported by Cynex Software.

Session S10

A QUADTREE-BASED EMBEDDED BLOCK CODER FOR SCALABLE IMAGE COMPRESSION

Chang-Mo Yang and Yo-Sung Ho

Kwangju Institute of Science and Technology (K-JIST)

1 Oryong-Dong Puk-Gu, Kwangju, 500-712, Korea

changmo@kjist.ac.kr, hoyo@kjist.ac.kr

Abstract: In this paper, we propose a simple but efficient wavelet-based embedded image coder that employs a modified quadtree partitioning. The proposed scheme includes multi-level dyadic wavelet decomposition, division of resolution layers, raster scanning within each sub-band, quadtree partitioning according to the parent-children relationship, and adaptive arithmetic entropy coding. Although the proposed scheme is simple, it produces a bitstream with a rich set of features, including resolution and SNR scalability together with the embedded nature. Experimental results demonstrate that the new scheme is quite competitive to and often outperforms other good image coders in the literature.

  OUTPUT FILTER UNIT IN DIGITAL VIDEO BROADCASTING TRANSMITTER'S CHAIN

Ivan Milak, Branka Zovko-Cihlar*, and Sonja Grgic*

R&D Department Hirschmann Austria GmbH, Ob. Paspelsweg 6-8,

A-6830 Rankweil, Austria

milak.ivan@rw.hirschmann.at

* University of Zagreb, Faculty of Electrical Engineering and Computing,

Department of Radiocommunications and Microwave Engineering,

Unska 3 / XII, HR-10000 Zagreb, Croatia

branka.zovko@fer.hr

Abstract: Intention of the paper is to show in reduced form the behaviour, choice and basic characteristics of the output filter unit in digital TV transmitter’s chain.

 MODELING AND SIMULATION OF MODERN COMPUTER NETWORKS

Daniel Z. Lenardić, Branka Zovko – Cihlar*

KPMG Consulting, *FER, University of Zagreb, Kaptol Centar, Nova Ves 11, Zagreb 10000, Croatia, Unska 3, Zagreb 10000, Croatia

e-mail for contact: daniel.lenardic@kpmg.hr

Abstract: This paper describes the problems, available tools, their capabilities and practical experiences involved in modeling and simulation of modern computer networks.

Session S11

LabVIEW SOFTWARE FOR A PARTIAL DISCHARGE TEST EQUIPMENT

S. D. Grigorescu, and V. Trusca

University POLITEHNICA of Bucharest Electrotechnical Faculty

Splaiul Independentei 313, 77206, Bucharest, Romania

e-mail: sgrig@electro.masuri.pub.ro

Abstract: This paper describes complete automatic test equipment for partial discharge in electric wires isolated with PVC. Instrument is a computer pending measurement and control system concerning high voltage control, high voltage coupling capacitor control, measure of additional and alarm parameters of high voltage power amplifier, test specimen automatic change and partial discharge measuring. Computer interface of the data acquisition and control system is designed in LabVIEW and has special aids for too slope high voltage test, cut ramp test, and measurements data base.

  STATE VARIABILE ACTIVE FILTER USING CURRENT CONVEYORS

Doru E. Tiliute Technical

University "Stefan cel Mare"
Suceava str. Mesteacanului No2, Bl 24, Sc A, apt 10, Suceava 5800, Romania,

e-mail: dtiliute@eed.usv.ro

Abstract: The author describes a method to achieve a new current-mode biquad active filter. The method is based bases on the transformation of the flow graph of a known conventional biquad active filter, as Tow- Thomas filter is. The resulted circuit has some useful features: it uses a reduced number of active and passive elements, all capacitors are grounded and that make it proper for integration, provides the orthogonal adjustment of Q, w o and gain and doesn’t require any match conditions to obtain the low-pass, high-pass and band-pass responses. At least, both input and output signals are voltages.

A DIAKOPTIC TECHNIQUE FOR SETTING UP THE SYMBOLIC STATE EQUATIONS OF LARGE-SCALE ANALOG CIRCUITS

Mihai Iordache and Lucia Dumitriu

Politehnica University of Bucharest, Electrical Engineering Department,

Spl. Independentei 313, Cod 77 206, Bucharest ROMANIA

Phone/Fax (+401) 411 11 90 iordache@hertz.pub.ro

Abstract: In this paper we present a decomposition technique to systematically formulate the state equations in symbolic normal-form for linear and/or nonlinear time-invariant large-scale analog circuits with excess elements. The decomposition and the aggregation procedures are shown and an illustrative example is given. The computation program allows the formulation of state equations in a symbolic form without any inverse of a symbolic matrix, providing a symbolic or partially symbolic compact form.


To the top