Very low bit rate face video compression using linear combination of 2D face views and principal components analysis

摘要

We present an image synthesis technique for very low bit rate encoding of faces for videoconferencing applications over the Internet or mobile communication networks. The proposed scheme represents each new face view as a linear combination of three basis face views of the same person. The background of the target face view is also encoded by the linear combination method using the basis views used for the face area. However, changes in the eyes and mouth regions cannot be encoded by the linear combination scheme. These regions are encoded using principal components analysis with three training sets of eyes and mouth images of the same person. The scheme results in a very compact representation of the face video data. Good quality images are reconstructed with an estimated bit rate varying from 1384 to 1620 bits/frame.