Face Image Coding Using Face Prototyping

In this paper we present a novel approach for face image coding. The proposed method makes a use of the features of video encoders like motion prediction. At first encoder selects appropriate prototype from the database and warps it according to features of encoding face. Warped prototype is placed as first I frame. Encoding face is placed as second frame as P frame type. Information about features positions, color change, selected prototype and data flow of P frame will be sent to decoder. The condition is both encoder and decoder own the same database of prototypes. We have run experiment with H.264 video encoder and obtained results were compared to results achieved by JPEG and JPEG2000. Obtained results show that our approach is able to achieve 3 times lower bitrate and two times higher PSNR in comparison with JPEG. According to comparison with JPEG2000 the bitrate was very similar, but subjective quality achieved by proposed method is better.

• Jaroslav Polec
• Lenka Krulikovská
• Natália Helešová
• Tomáš Hirner

• Triangulation
• H.264
• Model-based coding
• Average face

[1] L. Debruine, B. Jones, FaceResearch.org. University of Aberdeen in
Scotland. 2010. http://www.faceresearch.org/demos/average
[2] P. Weiss, Coding of Facial Images by Triangulation, Thesis, SUT
Bratislava, 2010
[3] M. Turk, A. Pentland, "Eigenfaces for Recognition," Journal of
Cognitive Neuroscience, March 1991
[4] E. Šikudová, M. Gavrielides and I.Pitas, "Extracting semantic
information from art images," Computer Vision and Graphics :
International Conference, ICCVG 2004, pages 394-399, Warsaw, 2004.
[5] E. Šikudová, "Comparison of color spaces for face detection in digitized
paintings," Spring Conference on Computer Graphics SCCG 2007,
pages 135-140, Bratislava, 2007.
[6] Joint Photographic Experts Group, oficial webpage,
http://www.jpeg.org/jpeg
[7] Joint Photographic Experts Group, oficial webpage,
http://www.jpeg.org/jpeg2000
[8] J. Polec, J. et al., Selected Methods for Data Compression, Bratislava:
UK, 2000. 196 p.
[9] Image and Video Compression Learning Tool VcDemo, Signal &
Information Processing Lab, Delft University of Technology,
http://siplab.tudelft.nl/content/image-and-video-compression-learningtool-
vcdemo