- Description of the Research Group +
The research at the Signal and Image Processing Group (SIPg, http://sipg.isr.ist.utl.pt) develops in four main topics related to information processing from sensory data: Sensor and Information Networks, Ocean Acoustics, Image and Video Analysis and Biomedical Engineering.
Its consistency stems mostly from a common attachment to designing algorithms/mathematical tools for data processing anchored on a shared methodological ground. To the more thematic character of these areas, SIPg adds a horizontal area named "fundamentals". The blueprint of this structure will become evident ahead.
Fundamentals -This area comprises work deeply rooted in mathematical disciplines like optimization theory, differential geometry, linear algebra and stochastic processes, in order to create and extend the "conventional" signal processing techniques to a new level. Examples are matrix completion under rank constraints, that arise in computer vision to handle occlusions in multiview settings or in localization and target tracking in sensor networks, and combinatorial problems such as bipartite graph matching, arising in several machine learning problems and optimization algorithms on manifolds which enable the development of information processing systems for high-dimensional data.
Sensor Networks - Sensor Networks are a "technology" for today's interconnected and dynamic society. From energy grids to transportation, the environment or physical infrastructures, sensor networks play a significant role in devising "ubiquitous" sensing and actuation mechanisms. The main challenges and virtues hinge on the intrinsic distributed nature of the data collection/processing, the randomness in the connectivity coupled with intrinsic constraints like bandwidth or energy. In this area we focus on the study of information processing in networked environments, enabling a wide range of applications involving massive sensing and/or "BIGDATA" contexts. Specific applications have been in energy efficiency in buildings, pollution monitoring in urban environments and target localization and tracking for mobility.
Image and Video Analysis - Images and video signals contain huge amounts of information about the scene and the activities which are being watched. However, extracting information such as the shape and motion of objects moving in the scene or to recognize their class or type of activity is a very complex task. Work in this area has addressed specific challenges such as 3D reconstruction from large image sequences, object recognition from 2D and 3D data, image retrieval in large datasets, and activity recognition in networks of video cameras.
Biomedical Engineering -This is a growing area of research at SIPG. The work is mainly focused in Brain, Medical and Biological Image processing and Biomedical Signal processing (BSP) algorithms for Computer-Aided Diagnosis (CAD). Particular focus has been given to the early detection of Alzheimer's disease (MCI) from PET and MRI, carotid atherosclerocitc plaque characterization from ultrasound images, skin cancer detection from dermatoscopic images and segmentation and tracking of the heart from ultrasound images.
Ocean Acoustics -The development of technologies for the study of the underwater world is one of ISR's hallmarks and a quintessential environment for signal processing methodologies. SIPg has devoted a substantial effort to creating underwater communications infrastructures that provide networking (Internet) to underwater operations, where terrestrial communication technologies cannot be used. Furthermore, the activity in ocean acoustics comprises the development of algorithms for underwater localization, habitat mapping, bathymetry, underwater tomography, and mammal detection.
In the period of this report, SIPg published 79 articles in journals, 308 in conferences, had 4 European and 14 national projects running and completed 13 PhDs.
- Main achievements +
As stated earlier, SIPg's strategy stands on the strong interchange between the quest to answer a set of core "fundamental" questions and the need to develop advanced mathematical and computational tools to tackle the thematic research that interests the group.
One of such questions was "how to estimate missing entries of a matrix constrained to a fixed rank?". Variants of this mathematical problem spun several results with "twin" publications/contributions in mathematics and computer vision:
- A new result on spectrally optimal estimation of matrices with incomplete entries described in [J1,C1] proposes an optimal solution to the problem of 3D object reconstruction from images (partial views). It received the "DoCoMo USA Labs Innovative Paper Award" (best paper award IEEE-ICIP-2008).
To localize moving targets in sensor networks, the work in [J2] we proposed a method relying on techniques for completing matrices formed by range measurements between the target and a set of anchors (EDM's). Pinar O. Ekim's PhD thesis developed part of this research and received the IBM (Portugal) Science Prize 2012, the most prestigious scientific prize for ICT's in Portugal. SIPg holds four of these prizes (2002,2003,2009,2012).
The (low) rank constraint was also used to solve combinatorial problems involving permutation matrices. A new theoretical result proves unique solutions which led to a reliable image/object matching technique solvable by convex optimization [C2,J3].
Distributed signal processing is today an enabling methodology for several applications in large scale networked settings. A set of novel optimization algorithms were developed for intrinsically distributed settings with random failures of its links [J4]. Algorithms such as this and [J5] are key to enable applications in large (Internet scale) information processing tasks.
Biological and medical image processing we highlight methods to detect and track dynamic left ventricle (LV) contours of the heart from ultrasound images [J6,C3], for image reconstruction in Ultrasound and Fluorescence image denoising for chronic liver disease diagnosis and staging, and for skin lesion assessment and diagnosis[J7]. These methods are supporting clinicians and biologists in their activities, e.g., diagnosis of Alzheimer's Disease (AD) from PET images [C4,C5], assessment of the atherosclerotic disease of the carotid [C6], or help in melanoma detection from dermoscopic images [C7].
In [J8,B1] a unique underwater acoustic network is described. This is a major outcome of European Project UAN, led by SIPg's researchers. By transporting massive communication capabilities to ocean environments, where these are currently very limited, this achievement opens a new window on marine exploration activities, as well as on potential societal and commercial spillovers.
Project URBISNET (http://sipg.isr.ist.utl.pt/urbisnet) developed a network of mobile (bus-borne) pollution sensing platforms. It required the buildup of sensing hardware, communication infrastructure, localization technologies, and the development of city-wide atmospheric dispersion models [C8].
Finally, we highlight the number of European and National projects involving international and national long-standing partners such as IPATIMUP, NTNU or Carnegie Mellon University. and the fact that SIPg was able to attract a large number of PhD students in globally competitive grant calls. Of these, we single out 16 Dual PhD's with CMU and 3 with co-advisors in international/national cooperative endeavors. Along this path 4 US patents were submitted by group members through collaborative projects with Honda and internships at Google, Qualcomm and LucasFilm (not yet disclosed). Another one was submitted together with IPATIMUP researchers and very recently, SIPg takes part in the team of a prestigious ERC grant (FIELDS-KNOT) through the intervention of "our" mathematician Marko Sto¨ic.
- Structure of the Research Group +
SIPg is a large group with more than 90 members, if master's students and other collaborators are taken into account. It is a very dynamic group with great capacity to attract top talented students and researchers, motivated both for theoretical or experimental/applied challenges.
SIPg members are physically located in two sites: in Lisbon, a group composed mostly by IST faculty, and in Faro a group with members from University of Algarve. This geographical and institutional separation makes natural sharing the coordination of SIPg's activities, being Prof. Sérgio de Jesus of the Algarve team the deputy PI and Prof. João Paulo Costeira the group PI. Notwithstanding, interactions and research developments between members flow naturally, the role of the PI being to foster engagement of all members in the accomplishment of the group's strategy. A few important tasks are collectively assumed by SIPg's members. In particular the following critical ones:
- Objectives of the Research Group +
Like any other research group, or any human community for that matter, SIPg has a multitude of diverse objectives in sight. On the one hand we have personal scientific drives, and on the other hand we are committed to the mission that a research group must fulfill in a community, either in the organization we are affiliated to (LARSys, ISR), or in society at large.
SIPg's scientific standing derives from circumstances and from its geo-scientific positioning. Born as a Signal Processing group, by nature it is attracted to any sufficiently difficult and attractive challenge. Let us take the question posed by a former IEEE-SP society president ,
"So, putting it together, can we say that signal processing is an enabling technology that encompasses the fundamental theory, applications, algorithms, and implementations of processing or transferring information contained in many different physical, symbolic, or abstract formats broadly designated as signals and uses mathematical, statistical, computational, heuristic, and/or linguistic representations, formalisms, and techniques for representation, modeling, analysis, synthesis, discovery, recovery, sensing, acquisition, extraction, learning, security, or forensics?".
By exerting a little prudency, we will not dare to answer that question, but we will assume the author was implying that "yes" is a very likely outcome! Consequently, SIPg is a candidate to poke its nose into everything that "moves". In fact, from the founding moment in the (tele)communications domain, SIPg evolved into a group which spreads contributions in many areas hardly tied to the original keywords.
This is to say that we will keep investing on stretching the most "fundamental" signal processing questions because that is what gives us the innovative edge that allows such a wide and deep spectrum of "applicability". In the past this objective has been realized through leading scientific initiatives of some SIPg faculty, working in tandem with strategic hirings (one mathematician, for example), external collaborations and a very active talent recruiting strategy. To this end, available "structural" (non-project-attached) funding was absolutely key.
On the other hand, SIPg is committed to contributing with knowledge and technology to two very important societal pillars: sustainability and quality of life. These goals match perfectly with areas we have been working on, and also with the group composition/aims. Massive distributed sensing networks, information processing and decision in large scale-settings, visual processing understanding, biomedical signal/imaging processing and technologies for the ocean are key instruments for better management of cities, to provide affordable healthcare to its citizens, to improve our daily lives and, of critical importance today, to protect and properly manage our environment.
Contributions to LARSyS mission statement are obvious and SIPg can play a significant role in any collaborative initiative with all groups of LARSyS as well as with all thematic areas.
Our scientific drive requires action in three main directions which guide SIPg's initiatives:
a)Strengthening and fostering new partnerships with global players (International cooperation, European projects/programs, Carnegie-Mellon|Portugal Program, MIT-Portugal Program) b)Improving current and constructing new challenging testbeds (ex: pollution monitoring like URBISNET, ocean networking like UAN, energy-monitoring like NIP) c)Agressive talent recruiting of PhD students.
 José M. F. Moura, "What is signal processing?, President's Message," IEEE Signal Processing Magazine, Vol. 26, Issue # 6, Page: 6, November 2009; DoI: 10.1109/MSP.2009.934636.