Raphaele Balter

Contact Information

Email: raphaele (dot) balter (at) irisa (dot) fr Address: IRISA Campus Universitaire de Beaulieu 35042 Rennes Cedex - FRANCE Phone: +33 2.99.84.74.24 Fax: +33 2.99.84.71.71 Project Assistant: +33 2.99.84.72.52 (Stéphanie Lemaile) Office: C107 BLEU

Profile

In 2001 I received my Engineering Degree of the EPUN (Ecole Polytechnique de l'Université de Nantes) previously IRESTE and my Master Degree (DEA) in Automatic and Applied Computer Science of the University of Nantes.

Now I'm a PhD student in image processing and computer science. This PhD is a CIFRE convention between IRISA/TEMICS and France Telecom R&D, so I am also a research engineer of the IRIS lab of France Telecom R&D of Rennes.

I was involved in the ARC Telegeo dealing with 3D objects progressive transmission.

Research Activities

My works essentially deals with signal/image processing and computer science. The subject of my PhD Thesis is "Construction de modèles 3D maillés en vue de la navigation virtuelle à distance (Extraction of 3D mesh models for distant virtual navigation)". This work is under the supervision of Luce Morin (IRISA/Temics) and Patrick Gioia (France Telecom R&D). I have been working on several aspects of 3D model based representations of videos. The advantages of this type of approachs are many; the 3D is likely to be more compact than the pixel based representation, and once the scene transmitted, it can be used in 3D to perform free navigation or add virtual objects in the photo-realistic environment for augmented-reality applications.

Video representation based on a 3D model stream

Extracting a 3D models stream instead of a unique realistic model of the scene is an entirely automatic algorithm without assumptions on camera parameters, scene content or video length. This representation is particularly interesting but it also brings discontinuities at the transitions between the models. I thus tend to establish an evolving model.

• The first step to an evolving model is a post-treatment in order to link successive models together. I proposed several entirely automatic morphings, two dealing with regularly meshed depth maps and one working on 3D models, based on a parameterisation and a consistent global mesh.
• To work with a real evolving model, I also proposed a representation based on a 3D models stream with corresponding vertices and consistent connectivity.

3D model-based video coding

Contrary to state-of-the art video coding that relies on pixel-based prediction / correction paradigms, 3D model-based video coding aims to reconstruct the 3D environment captured and send it as such. Exploitation of particularities of the encoded content can dramatically reduce the size of its encoding by using specific video coders. While this approach has been widely studied in terms of modeling and acquisition, only few works have focused on compressing and streaming the resulting representation of the scene. Indeed, this aspect of the coding is crucial for broadcasting video or proposing virtual walk through, and should be done in a fully scalable way for displaying the content on arbitrary device, from the PC screen to the cell phone matrix In this context I work on an efficient and scalable coding scheme for transmitting a stream of 3D models extracted from a video.

• The proposed method is based on exploiting the interrelations existing between each type of information that have to be transmitted (3D, Texure and camera parameters), instead of coding them independently, allowing to reduce redundancy.
• Scalability is achieved through the use of wavelet-based representations for both texture and geometry of the models. A consistent connectivity is built for all 3D models extracted from the video sequence, which allows a more compact representation and straightforward geometric morphing between successive models. Furthermore this leads to a consistent wavelet decomposition for 3D models in the stream.

2D/3D Hybrid Modeling for videos

3D model extraction from a video gives a representation adapted to low bitrate coding ang provides enhanced fuctionnalities. But for pure rotation motion of the camera 3D information ca not be retrieved. We thus proposed an hybrid approach based on a stream of 3D models and on mosaics.