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 theoretic bounds for sequential massive random access to large database of correlated data

Participants : Thomas Maugey, Mai Quyen Pham, Aline Roumy.

Massive random access is a new source coding paradigm that we proposed. It allows us to extract arbitrary sources from an appropriately compressed database purely by bit extraction. We studied the sequential aspect of this problem where the clients successively access to one source after the other. Theoretical bounds have been derived, and it was shown that the extraction can be done at the same rate as if the database was decoded and the requested sources were re-encoded. As for the storage, a reasonable overhead is required. In [26], we derived the optimal storage and transmission rate regions to the case of more general sources, which occur in practical scenarios. For the lossless source coding problem, we considered non i.i.d. sources (i.e., with memory, but also non necessary ergodic). We also showed that, in the case source statistics are unknown, the rate is increased by a factor that vanishes as the length of the data goes to infinity. Lossy compression is another context of interest, in particular for the application to video. Therefore, we derived achievable storage and transmission rate regions under a distortion constraint for i.i.d. [26] and correlated [13] Gaussian sources. Similarly, the transmission rate-distortion region is the same as if re-encoding of the requested sources was allowed. We are currently extending this work, by studying the constraints of the successive user requests and their influence on the transmission-storage rates performance.

Correlation model selection for interactive video communication

Participants : Navid Mahmoudian Bidgoli, Thomas Maugey, Aline Roumy.

One application of the sequential massive random access problem is interactive video communication for multi-view videos. In this scheme, the server has to store the views as compactly as possible while allowing interactive navigation. Interactive navigation refers to the possibility for the user to select one view or a subset of views. To achieve this goal, the compression must be done using a model-based coding in which the correlation between the predicted view generated on the user side and the original view has to be modeled by a statistical distribution. A question of interest is therefore how to select a model among a candidate set of models that incurs the lowest extra rate cost to the system. To answer this question, one should evaluate the effect on the transmission rate of using at the decoder a wrong model distribution. This question is related to an open problem in information theory called the mismatch capacity. So, we did not tackle the question for any type of code as in the case of the mismatch capacity. In contrast, we focused on a type of code of practical interest: the linear codes. More precisely, we proposed a criterion to select the model when a linear block code is used for compression. We showed that, experimentally, the proposed bound is an accurate estimate of the effect of using a wrong model.

Compression of spatio-temporally correlated and massive georeferenced data

Participants : Thomas Maugey, Aline Roumy.

Another application of the sequential massive random access problem is interactive compression of spatio-temporally correlated sources. For example, highly instrumented smart cities are facing problems of management and storage of a large volume of data coming from an increasing number of sources. In [23] different compression schemes have been proposed that are able to exploit not only the temporal but also the spatial correlation between data sources. A special focus was made on a scheme where some sensors are used as references to predict the remaining sources. Finally, an adaptation of the scheme was proposed to offer interactivity and free selection of some sources by a client. This work was been done in collaboration with the Inria I4S project-team (A. Criniere), IFFSTAR (J. Dumoulin) and the L2S (M. Kieffer).

ICON 3D - Interactive COding for Navigation in 3D scenes

Participants : Navid Mahmoudian Bidgoli, Thomas Maugey.

In the context of the ICON3D project, in collaboration with I3S-Nice (F. Payan), we have proposed a novel prediction tool for improving the compression performance of texture atlases of 3D meshes. This algorithm, called Geometry-Aware (GA) intra coding, takes advantage of the topology of the associated 3D meshes, in order to reduce the redundancies in the texture map. For texture processing, the general concept of the conventional intra prediction, used in video compression, has been adapted to utilize neighboring information on the 3D surface. We have also studied how this prediction tool can be integrated into a complete coding solution. In particular, a new block scanning strategy, as well as a graph-based transform for residual coding have been proposed. Experimental results show that the knowledge of the mesh topology can significantly improve the compression efficiency of texture atlases.