2009/12/03 – Marco Cagnazzo : “Compression for the Network, Scalable and Multiple Description Coding”
| 3 décembre 2009 | ||
| 14:00 | au | 16:00 |
Séminaire INFRES TELECOM ParisTech
Organisateurs : Hugues Randriam & Dario Rossi
Speaker : Marco Cagnazzo (Télécom ParisTech)
Title : Compression for the Network: Scalable and Multiple Description Coding
Abstract : Traditional video compression techniques have the implicit assumptions that the final user characteristics and requirements are known and do not change with time. Moreover, usually robustness is not taken into account when designing a video coder. This approach is clearly unfit to the video delivery over computer networks.
In order to improve the adaption between the encoder, the network and the users, a departure from traditional approaches is needed. The basic idea is to split the video representation into sub-streams; in particular one can have a hierarchical organization of sub-streams or not. In the first case, called Scalable Video Coding (SVC), each new sub-stream refines the information provided by previous ones, but is useless if these are not received. In the second case, called Multiple Description Coding, any set of sub-streams is decodable, but the compression performances are degraded with respect to SVC.
In this talk, after motivating the interest for SVC and MDC, a brief recall about video coding techniques will be performed, followed by the description of main scalability techniques implemented into standards. In particular the trade-off between scalability, compression performance and complexity will be explored. Some attention will be given to the important “drift problem” for scalable video. Finally, a few words about non-standard scalable and MDC techniques will be given as well.
