organised by



» Home » Overview


Comments from Participants of MDA

Comment from Emanuele Salerno

Mass data analysis of images and signals

Emanuele Salerno

National Research Council of Italy, and ERCIM, MUSCLE Working Group

The seventh international conference on mass data analysis of images and signals, MDA 2012, has been held in Berlin from 13th to 20th July 2012, in conjunction with two other conferences, an industrial exhibition, and a number of tutorials and workshops. Two invited lectures were given by Xiaoqing Ding, of Tsinghua University in Beijing, IEEE and IAPR fellow, and by Petra Perner, IAPR fellow, director of IBAI in Leipzig, and MDA program committee chair. The former was on face recognition, the latter on pattern recognition and data mining applied to quantitative evaluation of cellular phenomena. The MDA 2102 Best Paper award has been granted to the contribution titled "A New Method for Multifractal Spectrum Estimation with Applications to Texture Description", by Jacques Lévy Véhel and Michel Tesmer, of the Regularity Team at INRIA Saclay and the MAS Laboratory at the Ecole Centrale Paris.

Never in History we have been so overwhelmed by data as we are today. However, data per se are not useful to us. Indeed, their usefulness only comes from their content of information, so, somewhat paradoxically, we would like to have as little data as possible, provided they contain the information we need. A single image is both a bunch of useless data and a goldmine of information. An entire database of signals and images is both a bunch and a goldmine at the maximum extent. But a mine is not useful per se. Extracting gold means sifting out most of the material.

This is why data mining is perhaps the most important toolkit in our society of knowledge. A toolkit we need to improve and enrich continuously, to discover unsuspected patterns and hidden veins of knowledge. All the contributions in this conference are examples of this ability, from discovering regularities inhuman faces or brain activity records, to finding suitable metrics in complex multivariate processes, such as the behavior of living cells, the output of a gene expression experiment, a speech signal, or the motion of multiple objects in a natural scene.

Also, signals and images are perhaps the best examples to justify our goldmine metaphor: spatial and temporal data are always so correlated that signals and images are always very large and highly redundant data sets. Whereas one of the tasks of data mining is to reduce redundancy, so as to limit the amount of data a human specialist is required to analyze, redundancy also means robustness, and another essential task of data mining is to exploit redundancy to produce robust results. The specialist needs to be confident that no useful information is lost in their reduced data sets. Automatic data mining and analysis must result in minimal sets of highly reliable, complete, robust, and easily accessible processed data. Especially when dealing with signals and images, this also tells us that imaging and visualization are essential in data mining.

Fundamental problems to solve, in order to mine, analyze, and deliver information from images and signals, are registration, segmentation, classification, categorization, recognition, retrieval, and compression. All of these have now been faced in some way, even though some of them have no general solution yet. The challenge posed by mass data, however, forces us to rely on new approaches enabling effective and practically applicable procedures to be developed. In this context, high-level approaches should gain importance over methods based on low-level features and, whereas statistics and computation are still essential, including semantics is becoming inescapable. Working on classes and categories through means such as ontologies, case-based reasoning, and other knowledgebased strategies can be the key to face the new challenges. This affects equally the development of theory, the design of new methods and the practical applications. And this is why MDA and the associated events are devoted to both the research and the industry worlds.


Sponsors: Imageinterpret GmbH, Imaging & Microscopy magazine, ERCIM Working Group on Multimedia Understanding through Semantics, Computation and Learning (MUSCLE), Ganymed, Reisswolf.

Proceedings in Petra Perner, Ed., Advances in Mass Data Analysis of Images and Signals in Medicine, Biotechnology, Chemistry and Food Industry, ibai-publishing, Fockendorf, Germany, 2012, ISBN 978-3-942952-15-6.

Next appointment with MDA in New York, 13th to 16th July 2013.

Download as PDF