Full-frame video coding for facial video sequences based on principal component analysis
Ulrik Söderström, Haibo Li · 2005
Facial video sequences are encoded at very low bitrates through the use of principal component analysis. All facial emotions can be generated by blending six basic emotions. Principal components are used to model the basic facial mimic and encode facial video frames into a few parameters, which can be transmitted over low capacity networks. Hands-free video is used to circumvent large normalization, enabling realtime performance. From a bitstream of 2.4 kbps, video with high framerate, spatial resolution and visual quality can be reconstructed with an average PSNR above 31 dB. Encoding of larger frame sizes only increases the processing time, the bitstream is kept at the same low level. The effect of using shifting backgrounds is examined and this only have a moderate effect on the quality measured in PSNR. keywords: very low bitrate video coding, video compression, facial mimic, principal component analysis 1.