Previous | 
Home
Section: Bilateral Contracts and Grants with Industry
Bilateral Contracts with Industry
Contract with Astrium on compression of satellite images
Participants : Jeremy Aghaei Mazaheri, Christine Guillemot, Claude Labit.
-
Research axis : § 6.3.3 .
This contract with Astrium addresses the problem of sparse representation and dictionary learning for efficient sparse coding of video signals captured from a geostationary satellite. The goal is to develop a compact spatio-temporal representation taking advantage of the high redundancy present in the video which is of very high resolution and characterized by low motion. Different methods for learning tree-structured dictionaries have been studied. The tree-structured dictionaries are well-tailored to the characteristics of the signals to be processed at each iteration of the greedy matching pursuit algorithms, while allowing efficient encoding of the produced sparse vectors. Adaptive tree-structures have been developed and the use of such dictionaries in HEVC-based intra coding has been investigated. First tests have also been carried out to known to which extent the learned dictionnaries can allow detecting the modulation transfer function (MTF) used to characterize the quality of electro-optical imaging systems on board remote sensing satellites.
Collaboration with Alcatel on robust video compression
Participants : Marco Bevilacqua, Christine Guillemot, Ronan Le Boulch, Aline Roumy.
-
Research axis: 6.2.3
-
Funding: Joint research laboratory between Inria and Alcatel
In the framework of the joint research lab between Alcatel-Lucent and Inria, we participate in the ADR (action de recherche) Selfnets (or Self optimizing wireless networks). The objective is, jointly with the Alcatel Lucent team, to develop video representations and compression tools allowing smooth network adaptation on one hand and loss resilience on the other hand. In that context, the PhD thesis of M. Bevilacqua focuses on the development and study of image and video super-resolution as a tool for constructing scalable representations, hence enabling network adaptation of transmitted video streams. Single-image super-resolution algorithms have been developed, using different methods (neighbor embedding, local learning with regression), and dictionaries learned from external training images or learned on a multi-resolution pyramid constructed from the input low resolution image. These methods have been extended to video super-resolution, the dictionary being constructed from key frames.
Contract with EutelSat on video traffic analysis
Participants : Laurent Guillo, Aline Roumy.
This contract with EutelSat (starting in August 2012) is a consulting contract and aims at analyzing the variation of the video traffic, when the video is encoded by HEVC. Indeed, the main characteristic of satellite broadcasting, as proposed by Eutelsat, is to provide a nearly constant video quality, which is obtained by variable video traffic (bit rate). Then, to address this variability issue, statistical multiplexing is used to share the resource among the users. However, statistical multiplexing needs a precise analysis of this variability. In this contract, we therefore analyze this variability, when the video is compressed with the upcoming video compression standard HEVC.
Contract with SHOM (Service Hydrographique et Océanographique de la Marine)
Participants : Alan Bourasseau, Olivier Le Meur.
The project consists in developing lossless and lossy compression algorithms for oceanographic data in partnership with ALYOTECH. The SIROCCO team contributes on the design and development of compression algorithms for this specific type of data, based on diffusion methods. The main constraint is the limited bandwidth used by the navy to transmit the data, i.e. an emitted message must be smaller than 4 kilo bytes. In 2013, the obtained quality versus rate performances has been assessed against those given by state of the art solutions (HEVC-Intra and JPEG-2000).