Research

LISA has developed expertise both in image analysis/pattern recognition and computer graphics. In the field of image analysis and pattern recognition, this unit develops new methods for object segmentation, recognition or tracking in 2D and 3D problems, multi-modal image registration, as well as statistical learning methods applied to image and data classification. Developed algorithms are related to biomedical, industrial and HMI (Human Machine Interface) applications.

The LISA image synthesis research activities focus on the end-to-end processing chain for Virtual and eXtended Reality (VR-XR). It involves capturing and free-viewpoint rendering of real-life scenes, including multi-camera acquisition and compression. At the processing level, we utilise specialized video representation formats; not only from conventional RGB, but also from depth and lenslet-based cameras. We also conduct research on the real-time synthesis of virtual viewpoints that were not previously captured by the cameras. Our expertise is further consolidated within the MPEG standardisation with focus on 3D and Immersion activities.

Most of the aforementioned 2D/3D audio-visual signal processing algorithms require heavy number crunching on large data sets and therefore need to rely on efficient multi-core parallelization to ensure low-latency, real-time processing. It is then necessary to consider the intricate relationship between the application requirements, the algorithmic structure and the architecture’s multilevel memory hierarchy. In particular, General Purpose GPU programming and its efficient partitioning into regular processing kernels with minimal data dependency crossovers call for a complementary expertise over the full application-algorithm-architecture value chain. In particular, recent Deep Learning developments for signal and image processing take advantage of the massively parallel computing power available on modern GPU cards; this approach is used in several PhD theses in the lab.

LISA, following a problem-centered approach, tackles all hardware and software aspects of the chain in multidisciplinary teams (engineers, computer scientists, MDs, biologists, jurists, archaeologists, artists ...) over multi-institutional collaborations to deliver functional applications. The research is funded both by institutional/public funds and industry collaborations. LISA's achievements include one patent, several highly cited biomedical papers, implementation of acquisition and thermoregulation devices for live cell imaging, multi-media event organization and international cultural heritage projects.