Tutorials

EUSIPCO-2008 is pleased to announce the following tutorials, which will take place on Monday, August 25th, 2008.

Distributed Source Coding: Theory and Practice

[slides in pdf]

Dr Vladimir Stankovic, Dept. of Electronic and Electrical Engineering, University of Strathclyde, Glasgow, U.K.
Dr Lina Stankovic, Dept. of Electronic and Electrical Engineering, University of Strathclyde, Glasgow, U.K.
Dr Samuel Cheng, School of Electrical and Computer Engineering, University of Oklahoma, Tulsa, USA

Monday August 25th, 8:30am - 11:50am

Distributed source coding (DSC) refers to separate compression and joint decompression of mutually correlated sources. One example is compression of multiple correlated sensor outputs that do not communicate with each other and the sensors send their compressed outputs for a centralized joint decoding. Though theoretical foundations were set more than thirty years ago, driven by applications such as wireless sensor networks, video surveillance, and multiview video, DSC has over the past few years become a very active research area with interest from both academia and industry. DSC has strong potentials to enable efficient and low-cost signal processing in sensor networks, to improve current video communication technologies and open the door for many exciting new multimedia applications. However, the impact of DSC is expected to be much broader and cannot be overstated, since the potential of DSC is limitless. The proposed tutorial addresses theory and application of DSC, with the following aims: 1) introduce the theory of DSC and its connections to multimedia signal processing and signal processing for communications, 2) survey recent advances and exciting progresses made in DSC, and 3) discuss open challenges and opportunities in both theory and practical DSC designs. We hope this tutorial will provide the impetus for more signal processing researchers to contribute to the exciting and challenging field of DSC.

Discrete Optimization in Vision and Graphics

 [slides in pdf] (part I)

 [slides in pdf] (part II)

Nikos Komodakis, Department of Computer Science, University of Crete
Ramin Zabih, Department of Computer Science, Cornell University

Monday August 25th, 8:30am - 11:50am

A wide variety of problems from Computer Vision and Graphics can be naturally formulated in terms of minimizing the energy of a discrete Markov Random Field (MRF), which is thus considered to be a task of fundamental importance for these fields. However, the resulting optimization problems are very challenging. On the one hand, this is due to that many MRFs exhibit a highly non-convex energy function that is NP-hard to optimize exactly. On the other hand, an additional difficulty comes from the fact that the MRFs encountered in practice are typically of very large scale, which further implies that computational efficiency should be very seriously taken into account while designing an MRF optimization method. It is exactly for the above reasons that MRF optimization has attracted a considerable amount of research over the last years. In this tutorial we will present an overview for some of the recent developments in the field of MRF optimization. To this end, we will review a wide range of state-of-the-art discrete optimization algorithms such as graph-cut based techniques, methods that rely on propagating messages in graphical models, as well as methods that are based on Linear Programming. The underlying ideas and principles behind all these methods will be explained, and their experimental performance will be compared. A number of important applications will also be discussed, drawn mainly from vision and graphics.


Noise Robust Speech Recognition

[slides in pdf]

Jasha Droppo, Microsoft Research, Redmond WA, 98052, USA

Monday August 25th, 8:30am - 11:50am

Any practical speech recognition system must be designed to be robust to both additive and convolutive noises. This tutorial will give participants a better understanding of the noise robustness problem, an awareness of both basic and advanced techniques, and a knowledge of current research trends. The course is presented in three parts. Part analyzes how cepstral features, and the Gaussian components used to model them, are both distorted by noise. Part two presents some relatively simple techniques that can be used to combat noise, and an comparative analysis of their merits. Part three introduces a sampling of recently developed techniques, including data-driven probabilistic approaches, speech enhancement for recognition, and advanced model adaptation.


Theory and Applications of Compressive Sensing

[slides in pdf]

Richard Baraniuk, Department of Electrical and Computer Engineering Rice University
Monday August 25th, 2:30pm - 5:50pm

Sensors, cameras, and imaging systems are under increasing pressure to accommodate ever larger and higher-dimensional data sets; ever faster capture, sampling, and processing rates; ever lower power consumption; communication over ever more difficult channels; and radically new sensing modalities. The foundation of today's digital data acquisition systems is the Shannon/Nyquist sampling theorem, which asserts that to avoid losing information when digitizing a signal or image, one must sample at least two times faster than the signal's bandwidth, at the so-called Nyquist rate. Unfortunately, the physical limitations of current sensing systems combined with inherently high Nyquist rates impose a performance brick wall to a large class of important and emerging applications. In digital image and video cameras, for instance, the Nyquist rate is so high that too many samples result, making compression by algorithm like JPEG or MPEG a necessity prior to storage or transmission. In imaging systems (medical scanners and radars) and high-speed analog-to-digital converters, increasing the sampling rate is very expensive or detrimental to a patient's health.

Compressive Sensing is a new approach to data acquisition in which analog signals are digitized for processing not via uniform sampling but via measurements using more general, even random, test functions. In stark contrast with conventional wisdom, the new theory asserts that one can combine "low-rate sampling" with digital computational power for efficient and accurate signal acquisition. Compressive sensing systems directly translate analog data into a compressed digital form; all we need to do is "decompress" the measured data through an optimization on a digital computer. The implications of compressive sensing are promising for many applications and enable the design of new kinds of analog-to-digital converters, cameras, and imaging systems.

This tutorial will overview the theory of compresive sensing, point out the important role played by the geometry of high-dimensional vector spaces, and discuss how the ideas can be applied in next-generation acquisition devices. Particular topics include sparse signal representations, convex optimization, random projections, the restricted isometry principle, the Johnson-Lindenstrauss lemma, Whitney's embedding theorem for manifolds, and applications to imaging systems, sensor networks, and analog-to-digital converters.


New Tools for Bayesian Inference: The Variational Approximation

[slides in pdf]

Nikolaos P. Galatsanos, Dept. of Electrical and Computer Engineering, University of Patras, Rion, Greece
Dimitris Tzikas, Dept. of Computer Science, University of Ioannina, Ioannina, GReece.

Monday August 25th, 2:30pm - 5:50pm

Bayesian inference is a powerful approach for modern statistical signal processing. The main drawback of this approach is that for complex generative models exact Bayesian inference is very hard because it requires calculation of a Bayesian integral (marginalization of the hidden variables) which in most cases of interest is intractable. A number of methodologies have been proposed to bypass this difficulty. The Laplace approximation is one such a methodology. However, in many cases it is inaccurate. Markov Chain Monte Carlo (MCMC) is another such methodology based on stochastic sampling that recently has been used extensively in signal processing problems. However, MCMC can be very demanding computationally and it is hard to establish convergence of the Markov Chain. A new methodology termed “variational Bayesian inference” has emerged recently in the Machine Learning community, and is gaining rapidly popularity in the signal processing community also. This methodology is based on a deterministic approximation of the posteriors of the hidden variables it is computationally tractable and in most cases gives rise to parameter updates in closed form. Furthermore, one can show that the popular expectation maximization (EM) algorithm is just a special case of this methodology. At first an introduction to the basics of Bayesian inference will be presented then the variational methodology will be addressed in an easy to follow manner. As illustrative examples of this methodology the linear regression and the Gaussian mixture modeling problems will be used. These problems were chosen because they are fundamental for many signal and image processing applications. Finally, we will conclude this tutorial by presenting examples of application of variational inference to specific signal and image processing areas.

Nikolaos P. Galatsanos received the Diploma of Electrical Engineering from the National Technical University of Athens in 1982. He then received an MSEE and Ph. D. degree from the Department of Electrical and Computer Engineering of the University of Wisconsin-Madison, USA, in 1984 and 1989, respectively. From 1989-2002 we on the faculty of the Department of Electrical and Computer Engineering of the Illinois Institute of Technology, Chicago, USA. From 2002-2008 he was on the faculty of the Computer Science Department at the University of Ioannina, Ioannina, Greece. Currently he is on the faculty of the Department of Electrical and Computer Engineering at the University of Patras, Rio, Greece. His research interests center on Bayesian methods for image processing and pattern recognition with emphasis in imaging for visual communications and medical applications. Dr. Galatsanos has served as associate editor for the IEEE Trans. on Image Processing, the IEEE Signal Processing Magazine, and the SPIE Journal of Electronic Imaging. Currently, he is an associate editor for the IEEE Signal Processing Letters. Dr. Galatsanos is a senior member of the IEEE and a member of the Technical Chamber of Greece.

Dimitris Tzikas received his B.S. and M.S. degrees in Computer Science from the University of Athens, and Ioannina, in 2002 and 2004, respectively. Currently he is completing his Ph. D. at the Department of Computer Science at the University of Ioannina. Mr. Tzikas is working on Bayesian and machine learning methods for signal and image processing problems.


Ambient Intelligent Media

[slides not available]

Artur Lugmayr, NAMU Lab., Tampere Univ. of Technology (TUT)

Monday August 25th, 2:30pm – 5:50pm

Media evolved from media that can be described as integrated presentation in one form (multimedia). From multimedia, media evolved towards embedding the consumer in a computer graphic generated synthetic world (virtual reality). From this point on, media evolved to the consumers directly exposed to the media in their natural environment, rather than computer interfaces (ambient media). In addition, media will be evolving towards a fully real/synthetic world undistinguishable from pure media integrating human capacity (biomedia or bio-multimedia) somewhere in the very far distant future.

The goal is to train and educate participants in new innovative service design for ambient multimedia. The course will cover potential and possibilities of this new multimedia field and its relation to other trends, such as ubicom, pervasive computation, affective computation, and tangible media. Specific key-concepts of ambient media are developed based on various business case studies.

The "ambient way of thinking" in media technology enriches the world of media by the following principles:

  • automation - media are aggregated smartly by systems;
  • natural interaction - humans interact intuitively and naturally;
  • proactive - systems know human desires and act on their behalf;
  • emotional - systems recognize and express human emotions;
  • transparency - transparent and augmented access to content;
  • ubiquitous/pervasive - hardware and software disappear;
  • beyond push/pull - systems aggregate content, rather than humans

The course will give a comprehensive overview of the key ideas behind ambient intelligence, business and consumer trends, underlying technology, and new forms of potentially emerging media types. The tutorial also covers latest trends such as new forms of interaction, user-created content, M2M collaboration, social aspects, business aspects, collaborative content, and location based services in the era of ambient intelligence.

The structure of the course is as follows:

  • Introduction of participants
  • Ambient intelligence - what is it?
  • History
  • State-of-the-art
  • Organizations and entities
  • General viewpoints
  • Business viewpoint
  • Consumer and social implications
  • First practical examples & use-cases

Basic concepts and technology of ambient media and introduction of the key-concepts

  • Introduction of the components of AmI
  • Hardware components (ambience)
    • Smart materials
    • MEMS & sensor technology
    • Ubiquitous communication
    • Adaptive software
    • Embedded devices
    • I/O device technology
  • Software components (intelligence)
    • Media management & handling
    • Emotional computation
    • Natural interaction
    • Context awareness
    • Computational intelligence
    • Presence technologies
  • Application Scenarios
    • Content production
    • Collaborative content
    • M2M interaction
    • Location based services
    • Smart home
    • Personalization

Dr Artur Lugmayr describes himself as a creative thinker and his scientific work is situated between art and science. His vision can be expressed as to create media experiences on future emerging media technology platforms. He is pursuing his second doctorate at the School of Motion Picture, TV and Production Design in Helsinki, Finland. He is the head and founder of the New AMbient MUltimedia (NAMU) research group at the Tampere University of Technology (Finland) which is part of the Finnish Academy Centre of Excellence of Signal Processing from 2006 to 2011. The research group focuses on the development of smart spaces for media. He is currently preparing his second, invidually authored, text-book entitled "Ambient Media and Beyond" with Springer-Verlag in 2006. He chaired the ISO/IEC ad-hoc group "MPEG-21 in broadcasting"; won the NOKIA Award of 2003 with the text book "Digital interactive TV and Metadata" published by Springer-Verlag in 2004; country representative of the Swan Lake Moving Image & Music Award; project proposal reviewer; invited key-note speaker for conferences; workshop organizer for conferences; reviewer for publications and book chapters; has contributed one book chapter and written over 25 scientific publications. He gained his scientific practical experience in Austria (University Linz, RISC), Finland (Tampere University of Technology, School of Motion Picture, TV, and Production Design) and Greece, where he participated in several research projects. He is the inventor of bio-multimedia - integrated human capacity and the MPEG-21 based Digital Broadcast Item Model (DBIM). His passion in private life is to be a notorious digital filmmaker. More about him at http://www.cs.tut.fi/~lartur and on http://www.cs.tut.fi/sgn/namu.

logo


General Chairman

J.-Ph. Thiran, EPFL, Switzerland
JP.Thiran@epfl.ch

General Co-Chairman
P. Vandergheynst, EPFL, Switzerland
Pierre.Vandergheynst@epfl.ch

Technical Program Co-Chairmen
P. Frossard, EPFL, Switzerland
Pascal.Frossard@epfl.ch

A. Cavallaro, Queen Mary,
University of London, UK
andrea.cavallaro@elec.qmul.ac.uk

Plenary Talks
M. Unser, EPFL, Switzerland
michael.unser@epfl.ch

S. Godsill, Cambridge Univ, UK
sjg@eng.cam.ac.uk

Special Sessions
C. De Vleeschouwer, UCL, Belgium
christophe.devleeschouwer@uclouvain.be

J. Louveaux, UCL, Belgium
jerome.louveaux@uclouvain.be

Tutorials
N. Paragios, Ecole Centrale, France
nikos.paragios@ecp.fr

F. Bimbot, IRISA, Rennes, France
Frederic.Bimbot@inria.fr

Publications
R. Reilly, UC Dublin, Ireland
richard.reilly@ucd.ie

Local Arrangements
M. Marion, EPFL, Switzerland
marianne.marion@epfl.ch

Finances
M. Bach Cuadra, EPFL, Switzerland
meritxell.bach@epfl.ch

Publicity
G. Olmo, Politecnico di Torino, Italy
gabriella.olmo@polito.it

M. Gabbouj, Tampere UT, Finland
Moncef.Gabbouj@tut.fi

International Liaisons
B. Macq, UCL, Belgium
macq@tele.ucl.ac.be

U.B. Desai,
Indian Inst. of Tech., India
ubdesai@ee.iitb.ac.in

D. Erdogmus,
Oregon H&S University, USA
derdogmus@ieee.org

Exhibits
J. Reichel, Spinetix S.A., Switzerland,
reichel@ieee.org

Copyright © 2007 Kenzan Technologies & Eusipco 2008