Research Axes

Based on the research trends in multimedia indexing and retrieval as well as the competence and interest of current partners, we are interesed in several research areas and directions for IPAL.


Medical Image indexing and Retrieval for Assisted Diagnosis, medical research and teaching

Keywords: Medical image quantification, segmentation and classification, Medical Image retrieval.
PI Singapore: Wee Kheng Leow (NUS)
PI France: Daniel Racoceanu (CNRS)

This research area tackle the problem of information access on a restricted domain: medical images with associated data. This domain is also related to the automatic analysis of digital medical image in order to extract meaningful information that can be queried. This research emphasis the merging of existing low level features extraction to enable image classification on this particular domain. We also explore the fusion of general image analysis techniques and highly dedicated medical image algorithm on some dedicated image modality, and some anatomy or pathology. Finally, we study the link between visual concept related to low level image features and user high level concepts.

Infosheet

Download the PDF
image of the infosheet

Related projects

MmedWeb SERC-A*STAR project
Onco-Media ICT-Asian Network
ISERE ITC-Asian Network


CMIMIA- Contextual Multimodal Interaction for Mobile Information Access

Keywords: context, image recognition, ambience intelligence, mixed reality, mobile HCI
PI Singapore: Joo Hwee Lim (A*STAR/I2R)
PI France: Mounir Mokhtari (TELECOM SudParis)

People are frequently on the move. With the rapid advances in small mobile communication devices and underlying infrastructure, ubiquitous information access, a key information seeking activity to gather relevant information required by a task anywhere and anytime, will becoming very important.
Existing Information Retrieval (IR) technologies (e.g. searching documents on the Web using Google), although they have given us more flexibility than the traditional database systems that focus on structured data, cannot provide good technical solutions for ubiquitous information access for three main reasons, described below respectively.
The context related to the task that yields the information need and seeking behavior is not formerly and systematically represented and utilized in information retrieval. For example, the current and intended location of a mobile user can reveal useful cue to reduce the search space and to improve the relevance of information. Information retrieval is not an end, but a means to satisfy the information need related to a particular task for a specific user under certain context. Hence, incorporating user, task, and context relevance is very challenging and important for useful mobile IR systems.
The development of mobile computing devices enables new type of information access. New mobile phones support recording and playing of images, video, and audio, besides the more common voice, scroll, and touch input. In particular, multimedia content captured at a remote location (e.g. image of a monument of interest) can serve as intuitive and rich query input for mobile information retrieval. This kind of natural input will help to alleviate the difficulty related to classical text query on small devices. On the other hand, multimodal query input (text or speech keywords, images etc) incorporating contextual information (time, location, sound, user identity etc) opens up exciting challenges and opportunities for IR research.
Interaction on a small mobile device is an integral and non-trivial aspect for mobile information access. Since both the query input and information retrieved are multimodal, the creation of an intuitive and flexible HCI (Human Computer Interface), taking into account the constraint imposed by small screen, device computing power, task and user relevance etc, deserves new thinking and experimentation. The selection and adaptation of information based on the content of retrieved information and context of query also present new opportunities for research to enhance user experience.
In short, IPAL believes that the above difficult technical issues have to be addressed together. Hence we propose to undertake this new and promising research area of Contextual Multimodal Interaction for Mobile Information Access (CMIMIA).

Infosheet

Download the PDF
image of the infosheet

Related projects

MoSAIC (ICT-Asia Regional Programme)


FermiCold- Quantum Degenerate Gases and Strongly Correlated Systems

Keywords: Bose-Einstein Condensates, Fermi Gases, Quantum Phase Transitions, Strongly Interacting Systems, Quantum Simulators, Quantum Information
PI Singapore: Berthold-Georg Englert (Prof, CQT, NUS)
PI France: Christian Miniatura (Research Director, CNRS)

FermiCold is one project under the collaboration initiated between the Centre for Quantum Technologies (CQT) at NUS and the CNRS (MPPU, ST2I). FermiCold aims at establishing a strong experimental and theoretical french-singaporean team (at the international level) in the field of quantum degenerate gases. An experiment operating with ultracold fermions and/or ultracold boson-fermion mixtures will be set up at CQT (start date January 2009) and the corresponding theoretical expertise will be developed. The scientific project aims at studying the physics of strongly interacting degenerate quantum gases loaded in two dimensional optical lattices and at understanding the interplay between quantum fluctuations and interactions, at the heart of the phenomenon of quantum phase transitions, together with the role of entanglement in these strongly correlated systems. As such these systems offer the unique opportunity to implement quantum simulators and could also be used as model systems to study some features of quantum information and quantum computation.


3D Objects Streaming over Best-Effort Networks

PI Singapore: Wei Tsang Ooi (NUS-SoC)
PI France: Géraldine Morin, Romulus Grigoras (A/Prof., CNRS-IRIT)

Advances in 3D scanning technology and mesh reconstruction algorithms have lowered the barrier in creating complex mesh objects and sharing them over the network. For instance, the Stanford’s Digital Michelangelo Project digitized statues made by Michelangelo and provided a software, ScanView, to allow users to remotely view the 3D version of the statues. Second Life, an online virtual world, allows sharing of user-created 3D objects. Users download these 3D objects on-demand as they explore the virtual world. These signs suggest that an increasing amount of 3D mesh data will be available for remote viewing over the Internet.
The amount of data constituting a high quality 3D mesh can be huge. For example, the statue of David, from the Digital Michelangelo Project, consists of 2 billion polygons and constitute 32 GB of data after compression. To reduce waiting time when downloading such a huge amount of data, a common technique for remote viewing is to encode a 3D mesh progressively , allowing a low resolution version of the mesh to be transmitted and rendered with lower latency. The refinement information is continuously being transmitted, and the quality of the rendered model is incrementally improved over time.
This project aims to investigate issues in transmitting large scale objects and scenes over a lossy network such as Internet. Specifically, we aim to improve the rendered quality of the scene by appropriately packetizing (grouping 3D data into packets) and scheduling (when to send which packets) at the sender. Our approach will be based on rigorous analytical modeling of the effects of networks dynamics on the quality of the 3D objects. The team behind this project has worked on the problem for a year. We current plan to extend and generalize the model, to include different types of objects and dynamic scenes, and develop a showcase application, which will act as a proof of concept as well as experimental platforms to evaluate the effectiveness of the proposed methods.