Research direction 1 – Visual data analysis and visual attention modeling
Most visual data processing problems requirea prior step of data analysis, of discovery and m odelling of correlation structures. This is a pre-requisite for the design of dimensionality reduction methods, of compact representations and of fast processing techniques. These correlation structures often depend on the scene and the acquisition system. Scene analysis and modelling from the data at hand is hence also part of our activities. To give examples, scene depth and scene flow estimation is a cornerstone of many
approaches in multi-view and light field processing. The information on scene geometry helps
constructing representations of reduced dimension for efficient (e.g. in interactive time) processing
of new imaging modalities (e.g. light fields or 360° videos). Considering the human visual
system, its capacity of data analysis is related to visual attention, i.e. to the mechanism allowing
us to focus our visual processing resources on behaviorally relevant visual information. One goal
of our research is also the understanding and modeling of overt attention, involving eye mouvements,
as steps towards bio-inspired visual data processing.
Research direction 2 – Signal processing and learning methods for visual data representation and compression
Sparse, low rank models, graph-based models and processing, dimensionality reduction techniques
have been at the core of signal and image processing methods, for a number of years
now, hence have obviously always been central to the research of Sirocco. The study of these
models is becoming even more compelling for designing efficient algorithms for processing the
large volumes of high-dimensionality data produced by novel imaging modalities. They need to
be adapted to the data at hand through learning of dictionaries or of neural networks. The learning
of low-dimensional local models requires taking into account underlying data geometry or scene
Research direction 3 – Algorithms for inverse problems in advanced visual data processing
Our goal is to develop algorithms based on the above models for addressing various processing pr oblems, compression of course but also for solving a number of inverse problems in computer vision. Our emphasis is on methods to cope with limitations of sensors, for enhancing spatial, temporal resolution of capured data, for anti-aliased image-based rendering, for signal recovery from compression artefacts, or
for content editing. View synthesis is also central to that research axis for enabling multiview and
light field compression as well as user navigation and post-capture processing such as re-focusing.
Learning models for the data at hand is key for solving the above problems.
Research direction 4 – Coding for distributed processing and communication
The availability of wireless camera sensors has also been spurring interest for a variety of appl ications ranging from scene interpretation, object tracking and security environment monitoring. In such camera sensor networks, communication energy and bandwidth are scarce resources, motivating the search for new distributed image processing and coding solutions suitable for band and energy limited networking environments. Our goal is to address theoretical issues such as the problem of modeling the correlation channel between sources, and to practical coding solutions for distributed pocessing and communication.