Section: Overall Objectives

Distributed processing and robust communication

The goal is to develop theoretical and practical solutions for robust image and video transmission over heterogeneous and time-varying networks. The first objective is to construct coding tools that can adapt to heterogeneous networks. This includes the design of (i) sensing modules to measure network characteristics, of (ii) robust coding techniques and of (iii) error concealment methods for compensating for missing data at the decoder when erasures occur during the transmission. The first objective is thus to develop sensing and modeling methods which can recognize, model and predict the packets loss/delay end-to-end behaviour. Given the estimated and predicted network conditions, the objective is then to adapt the data coding, protection and transmission scheme. Classical protection methods use Forward Error Correction (FEC). The code rate is then adapted to the visual data priority. However, the reliability of the estimated PER, impacts the performance of FEC schemes. This is the first problem we propose to investigate focusing on the problem of constructing codes which, together with a scalable source representation, would be robust to channel uncertainty, i.e. which would perform well not only on a specific channel but also “universally”, hence reducing the need for a feedback channel. This would be a significant advantage compared with rateless codes such as fountain codes which require a feedback channel. Another problem which we address is the cliff effect from which suffer classical FEC schemes when the loss rate exceeds the error correction capacity of the code. The followed direction is based on Wyner-Ziv coding, used as a tool for lossy systematic error correction. The other problem addressed concerns error concealment. This refers to the problem of estimating lost symbols from the received ones by exploiting spatial and/or temporal correlation within the video signal. Classical approaches are based on spatial and/or spatio-temporal interpolation. We investigate new methods relying on video models (based on sparsity, epitomes, ...).

The availability of wireless camera sensors has also been spurring interest for a variety of applications 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 (Distributed Source Coding) solutions suitable for band and energy limited networking environments. In the past years, the team has developed a recognized expertise in the area of distributed source coding, which in theory allows for each sensor node to communicate losslessly at its conditional entropy rate without information exchange between the sensor nodes. However, distributed source coding (DSC) is still at the level of the proof of concept and many issues remain unresolved. The goal is thus to further address theoretical issues as the problem of modeling the correlation channel between sources, to further study the practicality of DSC in image coding and communication problems.