This thesis is devoted to the study of a compression and transmission framework for video. It exploits the JPEG 2000 standard and the coding with side information principles to enable an efficient interactive browsing of video sequences. During the last decade, we have witnessed an explosion of digital visual information as well as a significant diversification of visualization devices. In terms of viewing experience, many applications now enable users to interact with the content stored on a distant server. Pausing video sequences to observe details by zooming and panning or, at the opposite, browsing low resolutions of high quality HD videos are becoming common tasks. The video distribution framework envisioned in this thesis targets such devices and applications. Based on the conditional replenishment framework, the proposed system combines two complementary coding methods. The first one is JPEG 2000, a scalable and very efficient compression algorithm. The second method is based on the coding with side information paradigm. This technique is relatively novel in a video context, and has been adapted to the particular scalable image representation adopted in this work. Interestingly, it has been improved by integrating an image source model and by exploiting the temporal correlation inherent to the sequence. A particularity of this work is the emphasis on the system scalability as well as on the server complexity. The proposed browsing architecture can scale to handle large volumes of content and serve a possibly very large number of heterogeneous users. This is achieved by defining a scheduler that adapts its decisions to the channel conditions and to user requirements expressed in terms of computational capabilities and spatio-temporal interest. This scheduling is carried out in real-time at low computational cost and in a post-compression way, without re-encoding the sequences.

This thesis studies a neural network inspired by human neocortex. An extension of the recurrent and binary network proposed by Gripon and Berrou is given to store sparse messages. In this new version of the neural network, information is borne by graphical codewords (cliques) that use a fraction of the network available resources. These codewords can have different sizes that carry variable length information. We have examined this concept and computed the capacity limits on erasure correction as a function of error rate. These limits are compared with simulation results that are obtained from different experiment setups. We have finally studied the network under the formalism of information theory and established a connection between compressed sensing and the proposed network.

We are concerned with addressing the worst-case distributed source coding (DSC) problem in asymmetric and interactive communication scenarios and its application to data-gathering wireless sensor networks in enhancing their lifetime. First, we propose a unified canonical framework, obtained by considering different communication constraints and objectives, to address the variants of DSC problem. Second, as for the worst-case information-theoretic analysis, the notion of information entropy cannot be used, we propose information ambiguity, derive its various properties, and prove that it is a valid information measure. Third, for a few variants of our interest of DSC problem, we provide the communication protocols and prove their optimality. In a typical data-gathering sensor network, the base-station that wants to gather sensor data is often assumed to be much more resourceful with respect to energy, computation, and communication capabilities compared to sensor nodes. Therefore, we argue that in such networks, the base-station should bear the most of the burden of communication and computation in the network. Allowing the base-station and sensor nodes to interactively communicate with each other enables us to carry this out. Our definition of sensor network lifetime allows us to reduce the problem of maximizing the worst-case network lifetime to the problem of minimizing the number of bits communicated by the nodes in the worst-case, which is further reduced to the worst-case DSC problem in asymmetric and interactive communication scenarios, with the assumption that the base-station knows the support-set of sensor data. We demonstrate that the optimal solutions of the energy-oblivious DSC problem variants cannot be directly applied to the data-gathering sensor networks, as those may be inefficient in the energy-constrained sensor networks. We address a few energy-efficient variants of DSC problem and provide optimal communication protocols for the sensor networks, based on those variants. Finally, we combine distributed source coding with two other system level opportunities of channel coding and cooperative nature of the nodes to further enhance the lifetime of the sensor networks. We address various scenarios and demonstrate the dependence of the computational complexity of the network lifetime maximization problem on the complex interplay of above system-level opportunities.

No description available.