Side view driven facial video coding

Publicerad 2013 av Ulrik Söderström

Asymmetrical Principal Component Analysis (apca) Engineering And Technology Signalbehandling Teknik Och Teknologier Signal Processing Low Bitrate Facial Video Compression Low Complexity
Typ av publikation: Kapitel i bok eller rapport
Typ av innehåll: Övrigt vetenskaplig publ.

Ingår i:


A video can be decoded from an encoded version of another video when there is correspondence between the pixels in the two videos. Asymmetrical principal component analysis allows decoding of the frontal view of a face based on an encoded version the side view, or the profile of the side, of the same face. If the side view is used for encoding instead of the frontal view the reconstruction quality is only lowered slightly ($\approx$ 1,5 dB PSNR).

If the profile of the side of view is used to encode the video the complexity for the encoding is lowered with more than 99\% but the largest benefit of this coding is that the side view can be used for encoding instead of the frontal view. Filming the side view is far easier from a use-friendly point of view. This is a new kind of video coding where the information that is decoded does not have to be available for the encoding phase.