Self-Recovery Fragile Watermarking Scheme with Variable Watermark Payload

1
Self-Recovery Fragile Watermarking Scheme with
Variable Watermark Payload
Fan Chen, Hong-Jie He, Yaoran Huo, Hongxia
Wang Southwest Jiaotong University, Chengdu,
Sichuan, P.R. China
2

• Fan Chen
• Chengdu, Sichuan, China
• Southwest Jiaotong University
• An associate professor, and pursuing the Ph.D
  degree
• Research interests digital watermarking and
  Digital-forensics.

3
Presentation Outline
Self-Recovery Fragile Watermarking Scheme with
Variable Watermark Payload

1. Background
2. Proposed Self-recovery Scheme
3. Experimental Results
4. Conclusions and Future Work

4
1. Background

• Block codes
• 88 pixels DCT coefficients
• 22 pixels average intensity

Common Characteristic The length of block codes
is fixed.
Image Blocks
5
1. Background
Examples
Blocks
DCT quantized coefficients
Smooth
Rough
SIP laboratory
9. Z.Qian et al, Image self-embedding with
high-quaility restoration capability, Digital
Signal Processing, 21 (2011) 278-286.
6
1. Background

• Qian 9 proposed that the blocks were encoded
  to different number of bits.
• The rougher blocks have more bits
• And the smoother blocks have fewer bits.
• But, the amount of the compression code bits was
  still fixed for different images, no matter the
  image is smooth or rough.

9. Z.Qian et al, Image self-embedding with
high-quaility restoration capability, Digital
Signal Processing, 21 (2011) 278-286.
SIP laboratory
7
1. Background

• The watermark payload is more than the average
  length of the codes. This is due to the following
  two facts
• Tampering coincidence problem Dual watermark
  7, Reference Sharing Mechanism 2
• Tamper detection problem Authentication data

• Increasing watermark payload
• Decreasing the quality of watermarked image
• Being vulnerable to the collage attack

8
2 The Proposed Scheme

• Main purpose
• Reducing the watermark payload as much as
  possible while maintaining good recovery quality
  and security
• Strategies
• Variable watermark payload

9
2 The Proposed Scheme

• Watermarked Image Generation
• Variable-Length Block Coding
• Variable-Payload Watermark embedding
• Tamper Detection and Recovery
• Tamper detection
• Tamper recovery

10
2 The Proposed Scheme
-----watermarked Image Generation

• Variable-Length Block Coding
• For a block of 88 pixels
• The basic-code 24 bit
• The total-code the number of bits is
  variable

Code length (2427,24101)


Type s Type-code Code length vi (bits)
1 001 27
2 010 43
3 011 58
4 100 70
5 101 85
6 110 101
11
2 The Proposed Scheme
-----watermarked Image Generation

• Variable-Payload Watermark Insertion
• Watermark payload equals to the length of
  block-codes.
• Basic-code and total-code of a block are
  encrypted and hidden in the different blocks
  based on secret key.

12
2 The Proposed Scheme
-----Watermarked Image Generation

13
2 The Proposed Scheme
-----Watermarked Image Generation

• Tamper Detection Statistical Detection 5
• Basic-code mapping TB
• Total-code mapping TT
• Incorporation T

SIP laboratory
13
14
2 The Proposed Scheme
-----Watermarked Image Generation
Total-code matching
Detection results T
Basic-code matching
Tamper detection performance is better
5 He ea al, Adjacent-Block Based Statistical
Detection Method for Self-Embedding Watermarking
Techniques, 2009
SIP laboratory
14
15
2 The Proposed Scheme -----Tamper
Detection and Recovery

• Tamper Recovery Only for the invalid blocks.
• Watermark-reserved the tampered block whose
  basic-code or total-code is valid
• the corresponding codes
• Watermark-destroyed the tampered block whose
  basic-code and total-code are destroyed
• the valid blocks adjacent to the block

16
3. Experimental results

• The watermark payload and invisibility for
  different images

Images Watermark Payload (bpp) Proposed Qian9 Watermark Payload (bpp) Proposed Qian9 Invisibility (dB) Proposed Qian 9 Invisibility (dB) Proposed Qian 9
Boat 1.35 3 46.49 37.92
Lena 1.46 3 46.16 37.94
Airplane 1.47 3 46.02 37.93
Barbara 1.63 3 45.41 37.87
Mona Lisa 1.63 3 45.70 37.97
Elaine 1.64 3 45.52 37.94
Goldhill 1.73 3 45.12 37.93
Napoleon 1.74 3 45.06 37.96
Man 1.77 3 45.02 37.93
Fingerprint 1.95 3 44.36 37.79

17
3. Experimental results

• Restoration Performance

The tampered image can be recovered by the
proposed scheme with an acceptable visual quality
even the tamper ratio is up to 60 of the host
image.
18
3. Experimental results

• Collage attack

both of size 372288

(a) Watermarked Napoleon (b) Watermarked
Mona Lisa
19
3. Experimental results

• Collage attack

(d) Proposed 35.27 dB
(c) Tampered image (about 20)
20
3. Experimental results

• multi-region and multi-attack tampering

(1) General tampering (2) Only content tampering
(3) Collage attack (4) Constant-feature attack

21
3. Experimental results

(a) Proposed 34.38dB (b) Qian 9 16.58
dB (c) He 5 23.89 dB
Qian 9 and He 5 cannot resist all the
counterfeiting attacks
22
4. Conclusions and Future Work

• Our work
• A variable watermark payload
• Invisibility
• Recovery quality
• Security

23
4. Conclusions and Future Work

• Future work
• Extending this approach to resist mild
  distortion such as JPEG compression
• And analytic investigation on the tamper
  detection performance.

24
If you have any question, please e-mail
us. Email Fchen_at_swjtu.cn
hehojie_at_126.com

• Thanks a lot
• ??