Efficient face coding in video sequences combining adaptive principal component analysis and a hybrid codec approach

This paper presents a new proposal for a video coding scheme intended for human faces in video sequences using an eigenspace approach. The scheme greatly improves previous results by proposing a new adaptive eigenspace technique in combination with the new upcoming standard AVC. The description of the face is also included in the bit stream using the MPEG-7 descriptors thus providing search and browsing functionalities along with coding efficiency. The total bit stream presents the better or same coding efficiency than AVC/H.264 and offers a very competitive alternative to B-predictive frames. Results are provided in the paper for bit-rates around 2.5 Kbit/s.