Section: New Results

Distributed processing and robust communication

Information theory, stochastic modelling, robust detection, maximum likelihood estimation, generalized likelihood ratio test, error and erasure resilient coding and decoding, multiple description coding, Slepian-Wolf coding, Wyner-Ziv coding, information theory, MAC channels

Information theoretical bounds of Free-viewpoint TV

Participants : Thomas Maugey, Aline Roumy.

Free-viewpoint television FTV is a new system for viewing video where the user can choose its viewpoint freely and change it at anytime. The goal is to propose an immersive sensation without the disadvantage of Three-dimensional (3D) television (TV). Indeed, the conventional 3D displays (with or without glasses) occur, by construction, an accommodation-vergence conflict: since the eye tend to focus on the display screen (accommodation), whereas the brain perceives the depth of 3D images due to the different views seen by each eye (vergence). Instead, with FTV, a look-around effect is produced without any visual fatigue since the displayed images remain 2D. Therefore, FTV presents nice properties that makes it a serious competitor for 3DTV. Existing compression algorithms for FTV consider to send all the views, which would require about 100 Mbits/s (for 100 views, as needed to propose a true navigation within the scene). Since this amount does not fit the current datarate for transmission in a streaming scenario, we investigate a solution where the server only send the request. In [31] , [30] , we have shown a very surprising and positive result: if all the views are compressed once and if the server extracts from the compressed bitstream the request (i.e. one view at a time), the datarate is exactly the same as if the whole database was entirely decoded, and the requested views reencoded. This very positive result shows that it is possible to send FTV with the same datarate as single view television with very limited computational cost at the server (only extraction from the bistream). This result is an information theoretical result and the goal is now to build a practical system that can achieve this performance.

Compressed Sensing : a probabilistic analysis of the orthogonal matching pursuit algorithm

Participant : Aline Roumy.

Compressed sensing (CS) is an efficient acquisition scheme, where the data are projected onto a randomly chosen subspace to achieve data dimensionality reduction. The projected data are called measurements. The reconstruction is performed from these measurements, by solving underdetermined linear systems under a sparsity a priori constraint. It is generally believed that the greedy algorithm Orthogonal Matching pursuit performs well and can determine which variables are active (i.e. non zero). In contrast, we showed that this is not the case even in the noiseless context. We derived an exact probabilistic analysis of the iterative algorithm in the large system regime, when all dimensions tend to infinity. We showed that as the number of iterations grows, the algorithm will make errors with probability one.