Relations internationales :
Laboratoire Franco-Chinois d'Informatique,
d'Automatique et de Mathématiques Appliquées

organismes partenaires :
Académie des Sciences de Chine

Adresse :
Institut d'Automatique - Académie des Sciences de Chine
PO Box 2728 - Beijing 100080 R. P. Chine
Tél. : (+ 86 10) 62 64 74 59
Fax : (+ 86 10) 62 64 74 58

Site web en Chine : http://liama.ia.ac.cn/

L'Institut National de Recherche en Informatique et en Automatique (INRIA) et l'Académie des Sciences de Chine (CAS) ont créé à Pékin, en janvier 1997, un Laboratoire Franco-Chinois de Recherche en Informatique, Automatique et Mathématiques Appliquées (LIAMA), hébergé dès l'origine par l'Institut d'Automatique de l'Académie des Sciences de Chine (CASIA).

Depuis octobre 2004, le LIAMA compte sept partenaires :
· l'Académie des Sciences de Chine (CAS)
· l'Institut d'Automatique de l'Académie des Sciences de Chine (CASIA)
· le Bureau de Recherches Géologiques et Minières (BRGM)
· le Centre de Coopération Internationale en Recherche Agronomique pour le Développement (CIRAD)
· le Centre National de Recherche Scientifique (CNRS)
. l'Institut National de Recherche Agronomique (INRA)
· l'Institut National de Recherche en Informatique et en Automatique (INRIA)

Le LIAMA est une structure permanente de coopération. Ses missions sont :
· la conduite de projets de recherche associant des scientifiques chinois et français ;
· le développement de relations avec les communautés scientifiques et les industriels français ou chinois ;
· la formation, au travers des activités de recherche, d'étudiants et de spécialistes français et chinois.

Le LIAMA s'efforce également de faciliter la diffusion de logiciels issus des deux organismes vers l'industrie et les services utilisateurs et aussi de publier et diffuser des documents scientifiques dans ses domaines de compétence. Les aspects industriels jouent un rôle important dans cette collaboration.

Les principes de base de cette structure permanente de coopération internationale sont :
· la localisation au sein d'un seul organisme, l'Institut d'Automatique de l'Académie des Sciences de Chine à Pékin, de façon à servir de plate-forme aux communautés scientifiques des deux pays,
· le financement de projets conjoints sélectionnés par un comité d'experts bipartite, après appel à propositions,
· la présence d'équipes de recherche permanentes (projets résidents) hébergés dans les locaux du LIAMA,
· la valorisation d'applications industrielles et socio-économiques, notamment dans les domaines de l'environnement, de la santé et de l'aéronautique.

Les principaux projets rattachés au LIAMA sont :

Descritif des projets

Brain Imaging and Cognitive Disorders (BICD)
Principal Investigator : Dr. JIANG Tianzi (CASIA)

Brain Imaging and Cognitive Disorders is one team of LIAMA that brings together multidisciplinary expertise in Computer Science, Mathematics, Physics, Medical Imaging, and Neuroscience. It pursues a scientifically coherent program of internationally competitive research in Quantitative Imaging, Image and Signal Computing, and Medical Computer Vision - building on established strengths in these areas. It is keen on developing strong academic, clinical and industrial linkages, leading to significant domestic and international collaboration as well as the practical applications of its research.

Research themes include: Computational Neuroanatomy, Functional Brain Imaging, Imaging Genetics, and Complex Biological Networks, Systems Neuroscience, and Applications in Neurological and Psychiatric Diseases.

CAD : Computer Aided Design
Principal Investigators : Dr. Jean-Claude PAUL (INRIA) & Prof. Jun-Hai Yong, Director of Institute of Computer Aided Design, Tsinghua University

Computer Aided Design (CAD) Systems have a dramatic impact to day on the way designers and engineer’s work. In the industry, sketch and design, marketing, project review, pilot training, mechanical engineering, ergonomic studies or maintenance operations, all these works are based on numerical geometric models and simulations.

A class of surfaces is widely used in CAD. They are called NURBS that means “Non Uniform Rational B-splines”. This mathematical representation is very practical for Geometry Design. The designer can create curves with control points or weights very easily. However, many computation problems cannot be solved in a closed mathematical form with NURBS. For example, very often in Geometry Modeling, portions of an object are modeled independently with different B-spline surfaces that have different knock vectors. As a result, the B-spline surfaces do not match exactly. And the need of repair is very time consuming.

Triangle Meshes is another class of surfaces that is popular because they are supported by graphics hardware and computer games. We also have to deal with triangle surfaces when modeling is captured from real objects. However, the so-constructed mesh could be composed of a large number of elements. This makes it difficult to use it for many applications, like visualization. Applying numerical simulations is even more difficult, due to both performance and numerical stability problems. Clearly, triangle meshes appear to be a non-convenient mathematical representation

DeviceWare : Embedded Systems Software Development Tools
Principal Investigator : Dr. Vania JOLOBOFF (INRIA)

Embedded systems are more and more present in our everyday’s life, ranging from simple sensors to complex systems such as mobile phones, network routers, airplanes and defense apparatus. As embedded devices include increasingly sophisticated hardware, the development of software for embedded systems has become a key to economic development. The reader is referred to the European ARTEMIS R&D program for more complete analysis of the embedded systems software market analysis and research agenda.

We anticipate particularly on the following hypothesis in future embedded systems:
• Embedded target platforms will consist more and more ofr System On Chips integrating multiple processors and peripherals
• Hardware simulation at a high level of abstraction level, such as the TLM standard technology, or higher level, is going to become a key factor for gaining time to market in software development. Also as part of that, simulation environment will need to provide mixed simulated and real components.
• Virtualization technology is going to become more and more important in embedded systems development and will be more and more supported by hardware. Intel and AMD provide hardware support for virtualization on the PC, and virtualization is expected to become an important paradigm in the embedded market.
• Software development in general is going to be achieved within a model driven engineering approach, using a software component based approach.
• Embedded systems are less and less standalone systems, and more and more connected systems, whether loosely or strongly connected, and connectivity of the device is a growing factor of importance.
• Some of the embedded systems have to meet critical real time constraints, for example for quality of service, or linked to people safety (e.g automotive or aerospace industry). New issues in analyzing and proving real time or safety properties of the system entail from introducing virtualization technology and connectivity among the devices.

Eco-informatics : Plant Growth Modeling, Simulation and Visualisation
Principal investigator : HU Baogang (CASIA)

GREENLAB : Stochastic, functional and interactive models for plant growth and architecture
Coordinators : Baogang HU (CASIA) & Philippe de Reffye (CIRAD-INRIA)
GreenLab Research Issues
• Studies on the formalism and the behavior of GreenLab model based on instanciations to control the Plant Development.
• Interaction between Organogenesis and Photosynthesis through biomass production and partitioning .
• Computed from initial condition
• Deterministic and stochastic case
• Double reserve, dry and fresh matter (energetic mock-up)
• Organ functioning control step by step.
• Tree structure simplification.
• Interaction plant and environment: water, light , temperature (at organogenesis level)
• Optimization and Control of the dynamical growth to improve yield under constraints.
• Computer Graphics applications for plants and stands visualisation

Environmental modeling
Coordinators : Thierry Fourcaud (CIRAD) & Liang SHAO (Ecole Centrale de Lyon)
Understanding, modeling and simulating the main physical phenomena involved during environmental hazards must play a key role in the sustainable development of land in China and also in other countries. Environment research activities in LIAMA focus on studying the physical interactions between environmental components such as soil, plants and air flow. These studies are based on both experiments and development of mechanistic models that aim to simulate fundamental mechanisms and to predict the associated risks.
Research topics include:
• Turbulent dispersion of pollutants
• Eolian erosion and transportation
• Eco-engineering

IGIT : Interactive Image and Graphics Techniques
Principal investigators : Christophe CHAILLOU (INRIA) & PAN Chunhong (CASIA)

Our current research mainly focuses on human motion extraction and analysis, and 3D model editing and reusing.
In human motion extraction and analysis, based on the spatial and temporal information of human motion, we attempt to extract 3D motion of any part of human such as leg, arm from a monocular sequence image, also work on 2D tracking of human contour. Additionally, we also work on 3D motion of finger from a sequence image by making full use of the linkage kinematics of finger. The finger is modeled by 3D articulated model with different linkages. These results can be widely applied in 3D human-machine interaction.
In 3D model editing and reusing, we focus on 3D pose transfer between two different models. Given a source mesh and a target mesh, the pose of the source mesh can directly be transferred to the target mesh, while preserving the surface details of the target mesh. Different from the traditional deformation transfer method, our method does not require the reference source mesh. It also can successfully transfer both gross skeletal structure and subtle skin deformation.

Research Issues
• 3D reconstruction from images
• Sketch-based 3D mesh editing
• Parameterization and remeshing of 3D models
• 3D human motion editing and retargeting

NETQUEST, In-network route and query processing
Principal investigator : Dr. Stéphane GRUMBACH (INRIA)

Networks of independent entities, cooperating to handle global tasks, constitute a fascinating class of systems. They are widely found in nature, where cells exchange information with their neighbors, or in the brain with synapse connections between neurons, as well as at other levels of living organisms. They can be found also in various social organizations. Although these networks are now receiving considerable attention, their behavior is still poorly understood. Largely spread in nature, such organizations are making their ways into microelectronic systems. In the realm of information technology, ad hoc networks of autonomous devices might in the near future become ubiquitous. They are made possible by the development of ever smaller and cheaper electronic devices with increased memory capacity and computational power, together with the standardization efforts for both wireless communication and data exchange format. The main challenge of these often spontaneous networks is to solve global problems while relying only on local communications between their nodes. Indeed, their fundamental characteristic is that each node can communicate directly with only a restricted number of neighboring nodes. Communication between distant nodes can be performed by multihop passes in the network, but is limited by prohibitive communication cost.
The nodes involved in these networks have to support two fundamental tasks. (i) Networking: The nodes have to organize themselves to form a network, which can ensure efficient routing of the data over the network, and adapt to changes in the network topology due either to node failure or mobility. (ii) Querying: The nodes have to collaborate on query processing to identify the nodes storing relevant data, and distribute the computation of the queries over the devices, taking into account constraints on the devices as well as the bandwidth, while optimizing communication.
The objective of the NETQUEST project is to develop original solutions to support data intensive applications in dynamic networks. We focus on ad hoc networks of devices communicating with their neighbors with no centralized control. The novelty of our approach is to handle in a uniform fashion both the data and the network management, thus abolishing the frontier between network layer and application layer. This is achieved by expressing networking problems using declarative queries over the graph topology, much in the same way as queries over the distributed data. All queries, either on the data or on the network topology, are evaluated by a unique query engine, whose development constitutes the central objective of the project. The main R&D targets for the coming years are to develop the methodology and the algorithms for in-network combined query processing, the realization of a prototype of the query engine, its tuning over a simulation platform, its evaluation over a testbed of sensors, and its transfer to industry.

Pattern Analysis and Learning
Principal investigator : LIU Chenglin (CASIA)

The Pattern Analysis and Learning (PAL) Group was founded in Jan. 2005, by Dr. Cheng-Lin Liu, and supported by the “Hundred Talents Program” of CAS. The research topics of the group include theory and methods of pattern recognition and machine learning, and the applications to document analysis, image analysis and text categorization.
Pattern classification methods have been widely applied in practical recognition problems, but are desired to yield higher performance. Classification is also a main task in machine learning. The problems to study include parametric and non-parametric density estimation, dimensionality reduction and feature selection, hybrid generative/discriminative learning, hybrid statistical/structural classification, multiple classifiers generation and fusion, incremental learning, semi-supervised learning, etc.
Character recognition and document analysis techniques are aimed to process large volumn of paper and pen-based documents. The remaining difficult problems include the recognition of unconstrained handwritten characters, character segmentation, analysis of complex layout, character recognition in low-resolution or degraded images, and so on. Also, we need to upgrade the task of document analysis to the level of linguistic contents, i.e., it can be combined with text categorization and retrieval.

RSIU : Image Understanding for Remote Sensing Applications
Principal investigator : Dr. Véronique PRINET (CASIA)

The Remote Sensing Data Fusion and Understanding activity, as part of the Group "Visualisation and Remote Sensing" of the National Laboratory of Pattern Recognition (NLPR) was initiated in 2000. Since Oct.2004, this activity is also included with the Sino-French Research Laboratory of Information Automation and Applied Mathematics (LIAMA). The undertaken research works fall in the topics of image processing and understanding, from low to high level, with applications on environmental modeling in the broad sense and natural/man-made disasters.
Initialised with a first team of 3 persons, it grew up to 14 people in 2003. We are currently a team of 7, including two permanent researchers, three PhD candidates, two Master students. Every year, we welcome DEA or Engineering School internship students from France, within the context of the LIAMA.
The undergoing research contracts, mainly international cooperation projects, are supported by national and international organisms (EU-IST, EU-INCO, MOST-863, NSFC, WB, PRA, LIAMA).
We carry out a number of cooperation projects with French institutions. Among our academic partners in France are : INRIA-Ariana, Ecole Centrale, Univ.Rennes 1, ENSEEIHT, ENST. Our industrial partner is currently Alcatel-Aliena-Space.

Key words : Image processing, Image understanding, Object modeling and recognition, Knowledge representation, Invariant descriptors, Energy minimization approaches, Learning, Deformation/Motion/Change, Fusion, Earth observation, Remote sensing, Environment.

SCILAB Promotion in China
The Team : Dr. LI Shi, CASIA, SCILAB Promotion Manager & Dr. Claude GOMEZ, Coordinator in French side, INRIA/SCILAB Consortium

The objectives of this project are to promote SCILAB in China, which is the famous open source software for scientific computing co-developed by scientists of INRIA and ENPC. Since 1997, SCILAB has been extensively received by scientists, engineers and technicians, and practitioners in both academia and industry, which has been recognized as the best substitute for commercial software like Matlab, Xmath and Matrix, and distributed currently at the speed of 10,000 copies per month in all over the world.

SCILAB has brought us not only powerful functions in scientific computing, but also its free spirit of open source. The free spirit of open source software and technology makes far-reaching sense to China’s software industry in terms of liberalization from the international Commercial software companies. Therefore, research institutes and governmental sessions concerned in China and France have co-developed and been supporting SCILAB activities in China. From this point of view, SCILAB promotion in China setups an broad bridge between the researchers in China and in French, and provides the opportunities for Chinese researchers, engineers and students to study and to express their talent by means of competition and communication with industrial partners both in Chinese and French side.

In 2001, LIAMA organized the first Sino-French SCILAB workshop in Beijing, which was the starting point of a fruitful collaboration between LIAMA and INRIA. After that, the events became annual, and the contest Awards Ceremony followed by the Sino-French SCILAB workshop. It took place in Shanghai in 2002, in Xian in 2003, in Xiamen in 2004, in Wuhan in 2005 and Hangzhou in 2006. This contest is very important because it makes SCILAB known in Chinese Universities and it provides SCILAB with good quality toolboxes.
More information: http://www.scilab.org.cn/