User adaptive transcoding for video teleconferencing

The human visual system (HVS) cannot perceive spatial frequency components in an image that are above a certain limit, the value of which is affected by factors such as the viewing distance. This has been exploited to improve the video coding efficiency by first filtering out redundant frequency components and then doing conventional video encoding. To facilitate the deployment of this scheme, we propose to implement it in a network entity such as a Multipoint Control Unit (MCU). Specifically, by analyzing the video sent from a client, the MCU infers that client's viewing conditions, which are then used to adapt the encoding of the video destined to that client. The scheme is implemented in a real-world video teleconferencing system. Experimental results show that our approach can result in significant savings in bandwidth without affecting subjective video quality.