Iterative Frame Decimation and Watermarking for Human Motion Animation

1
Iterative Frame Decimation and Watermarking for
Human Motion Animation

• Shiyu Li, Masahiro Okuda

Faculty of Environmental Engineering, The
University of Kitakyushu 1-1,Hibikino,
Wakamatsu-ku, Kitakyushu, Fukuoka,
Japan lishiyu, okuda-m_at_env.kitakyu-u.ac.jp http
//vig.is.env.kitakyu-u.ac.jp/
2
PRELIMINARY
Body Skeleton Configuration
3
Introduction- Why Motion Coding?

• Reuse motion data (storing)
• Motion data retrieval
• Motion editing
• Real time transmitting

4
Introduction- Two methods

• Reduce motion sample density
• -key-frame, interpolation
• Reduce size of database
• -compression of each sample itself

5
Reduce motion sample density

• key-framing
• Shiyu Li, Masahiro Okuda and Shin-ichi
  Takahashi, Embedded Key-frame Extraction for CG
  Animation by Frame Decimation, Proceeding of
  IEEE Int. Conference on Multimedia Expo
  (ICME05)
• Watermarking
• Shiyu Li, Masahiro Okuda, Watermark for
  Progressive Human Motion Animation, Proceeding
  of IEEE Int. Conference on Multimedia Expo
  (ICME07)

6
Reduce size of database

• Wavelet transform based
• Shiyu Li, Masahiro Okuda, Shin-ichi Takahashi,
  Hierarchical Human Motion Compression with
  Constraints on Frames, 47th MWSCAS, Hiroshima,
  Japan, pp. I-253- I-256, Jul. 2004.
• Using the Reduction of Inter-Joint Correlation
• Shiyu Li, Masahiro Okuda, Shin-ichi Takahashi,
  Kinematics-based motion compression for human
  figure animation, 30th ICASSP, Pennsylvania,
  United States, vol.II, pp.1077-1080, Mar. 2005.

7
Embedded Key-frame Extraction for CG Animation by
Frame Decimation

• Previous method motion curve simplification
• Our method All the frames are ranked by their
  importance, and thus users can specify any number
  of key-frames from one data set.

8
Watermark for Progressive Human Motion Animation

• A progressive coding based watermarking scheme

To embed the watermark in the encoder The
sequence E is broken up into blocks. We use 50
frames per block. To calculate the variance
Var(E) of the signal E in each block (1). if the
value of a frame is larger than Et, Et is
subtracted from this frame. Fig.2(b) represents
the signal after this preprocessing. (2). to
produce a sequence which is zero-mean, the mean
of each block is subtracted from each frame.
Calculate the variance Var(E) of each
block. Calculate the variance Var(w) of the
watermark w by the Var(E). Generate a pseudo
random sequence w with the same length as the
blocks of sequence E, assume E and w are
independent. See Fig2.(c) Store EEw as the
watermarked signal. See Fig2.(d) The watermarked
signal E is sent to the decoder. In the
decoder Rank the frames by the value of
E. Receive the key-frames progressively as this
order. After applying interpolation to the
key-frames, the motion can be reconstructed
approximatively. To detect if the watermark
exists, the original Et is necessary Calculate
the variance of E after the same preprocessing
in the encoder, i.e. if the value of a frame is
larger than Et, Et is subtracted from this
frame. Estimate the variance of w, assuming
Var(E)Var(E) Var(w). Var(E) is calculated by
the same method in step1. Regenerate w using the
same seed value. Calculate ZS(w E) Declare E
watermarked if Z gt n Var(w) / 2
9
Hierarchical Human Motion Compression with
Constraints on Frames
10
Kinematics-based motion compression for human
figure animation
Endcoder
Decoder