Joint Data Hiding

1
Joint Data Hiding Source Coding with Partially
Available Side Information

• Ç. Dikici, K. Idrissi, A.Baskurt

LIRIS, UMR 5205 CNRS INSA de Lyon, France
2
Agenda

• Data Hiding and Distributed Source Coding
• System Architecture
• Experiments
• Conclusion

3
Motivation

• Coding of multiple sources which can not
  communicate at the encoder
• Ultra-low encoding complexity
• Reduce power consumption
• Data hiding for each source signal

4
Literature

• Wyner-Ziv, Gel'fand-Pinsker
• Ramchandran et al,  Duality between source
  coding and channel coding with SI , Asilomar99,
  ISIT00, 01
• Girod et al.,  Duality between Channel Coding
  and Rate distortion with SI , Asilomar00
• Cover et al.,  Duality Between Channel capacity
  and Rate distortion with two sided State
  Information , ITT02

5
Theory(1/3)

• Source Coding with SI (R01)
• Rate

6
Theory(2/3)

• Channel Coding with SI (Blind watermarking or
  C10)
• Capacity

e
Y
W
Encoder or quantizer
Decoder
M watermark
M(Y)
Z (attack)
S1
7
Theory(3/3)

• Channel Coding with SI (General solution
  Cgeneral)
• Capacity

W
e
Y
Encoder or quantizer
Decoder
M watermark
M(Y,S2)
Z (attack)
S2
S1
8
Proposed schema
Outter C
Inner R01
Receiver
Sender
coset C
CCSI Encoder
SCSI Encoder
SCSI Decoder
e
CCSI Decoder
W
W
M
M
S1
S2
S1,S2 p(S1,S2) S1,W p(S1,W)
9
Experimental Setup
W
P

• DATA HIDING
• Embed 2 bits of M within R bits of S1
• Using
• coset1000,111
• coset2001,110
• coset3010,101
• coset4100,011
• M3
• Slsb100
• gtW101

• INPUTS
• S1 100x2000 bit blocks with Bernoulli(1/2)
• S2 S1N where
• N Bernoulli(p)
• M Random message

• ENCODING
• 3/5 LDPC code
• with H(3500,2000)
• parity check matrix
• send only parity bits

• DECODING
• Use S2 and P
• Decode with modified Belief propagation

• Estimation
• Final estimate of W and M

W
Rate r encoder
S2
Decoder
W
M
10
System performance
Pe0 for plt0.13
Channel capacity for rate 3/5 is pe0.08 Since
errors are only in systematic bits 0.133/5
0.078
11
Conclusions

• A DSC system with data hiding
• Works well up to a certain correlation to noise
  ratio
• On sensor networks, each client can embed its
  specific data to the stream
• Will be extended to the multimedia data like
  image, and video

12
Thank you for your attention
13
System realization
Compression params.
Encoder
Decoder
Output statistics
Virtual channel