Ph.D. student under the direction of F. DUFAUX

Thesis title: Video for Events: Compression and Transmissionof the Next Generation Video Codec
Thesis abstract: live video streaming has undergone a revolution in recent years with the emergence of applications allowing everyone to broadcast their activities live on social networks, either for personal or Professional purposes, and with very variable qualities. The ISO / IEC MPEG and ITU-T VCEG standardization groups began exploratory work in 2015 on new coding tools that already allow a performance gain of 25% to 30% compared to HEVC. This motivated the launch of a formal standardization activity in order to achieve a 50% coding gain compared to HEVC by 2021. This new standard has objectives related to UHD (4K and 8K). The resources in the event are however more limited in terms of bandwidth, with rates ranging from 128kbp s for degraded satellite links to around 10Mbp s for 4G networks. These use cases have also low latency requirements. In addition, depending on the type of event, accessibility to network resources can vary widely. This thesis fits in this context. The objective is to study the compression tools in contexts of lower resolution and at rates lower than their original constraint, and to adapt and optimize the video codec in these cases of use and therefore, increase the coding efficiency compared to the state of the art. Then, adapt the average quality of the encoded video according to the content and characteristics of the transmission channel. Thus, we want to study the control of the coding parameters and the level of the transmission buffer, and create a model that allows the prediction of the channel characteristics. Statistical learning mechanisms will be used to improve the quality of prediction. The next step is to study the impact of this predicted informations on the optimization of the coding parameters. Parameter selection techniques based on Markovian Decision Process will be used. Index Terms: video compression, real time adaptive streaming, adaptive video coding

live video streaming has undergone a revolution in recent years with the emergence of applications allowing everyone to broadcast their activities live on social networks, either for personal or Professional purposes, and with very variable qualities.

The ISO / IEC MPEG and ITU-T VCEG standardization groups began exploratory work in 2015 on new coding tools that already allow a performance gain of 25% to 30% compared to HEVC. This motivated the launch of a formal standardization activity in order to achieve a 50% coding gain compared to HEVC by 2021. This new standard has objectives related to UHD (4K and 8K). The resources in the event are however more limited in terms of bandwidth, with rates ranging from 128kbp s for degraded satellite links to around 10Mbp s for 4G networks. These use cases have also low latency requirements. In addition, depending on the type of event, accessibility to network resources can vary widely.

This thesis fits in this context. The objective is to study the compression tools in contexts of lower resolution and at rates lower than their original constraint, and to adapt and optimize the video codec in these cases of use and therefore, increase the coding efficiency compared to the state of the art. Then, adapt the average quality of the encoded video according to the content and characteristics of the transmission channel. Thus, we want to study the control of the coding parameters and the level of the transmission buffer, and create a model that allows the prediction of the channel characteristics. Statistical learning mechanisms will be used to improve the quality of prediction. The next step is to study the impact of this predicted informations on the optimization of the coding parameters. Parameter selection techniques based on Markovian Decision Process will be used.

Index Terms: video compression, real time adaptive streaming, adaptive video coding