Semifragile Watermarking

Technical Seminar on Comparative Evaluation of Semifragile Watermarking Algorithms

Presented by: Rajesh S.Hegde 4JN05IS042

Contents: 1.

Introduction

2.

Methods for Comparison

3.

Semifragile Watermarking Algorithms

4.

Experimental Results

5.

Conclusion

6.

References

Introduction: 

Semifragile watermark monitors the integrity of the content of the image but not its numerical representation.



Prominent application of watermarking technology is monitoring of the integrity of multimedia documents.

 Attacks on image can occur as follows, iii.Unintentional

or Innocent alterations: Arise as bit errors during transmission and storage or signal processing operations such as filtering, contrast enhancement, sharpening and compression. ii.

Intentional or Malicious alterations: Due to an explicit forgery attempt by a pirate with an explicit purpose of changing the contents of a document

 Watermarking algorithms: iii.Fragile

watermarking algorithms:

Function

as strict tamper detection tool and intolerant of even one bit alteration. Also called as cryptographic message digests and can validate only original copies. viii.Semifragile Monitor

Tamper detection methods:

changes in the content. Capable to differentiate between innocent and malicious attack categories.

 Classification of tamper detection content authentication watermarking techniques: iv.Visually

authenticated (semi)fragile watermarking:

Typically

thumbnail or a visual pattern is hidden in the

image. Tamper detection is based on visual assessment of perceived differences by an operator.

ii. Statistically authenticated (semi)fragile watermarking An

estimate of tampering likelihood is obtained based on correlation coefficient Or Measured mismatch between inserted and recovered authentication sequences.

Proposed algorithm may be, i. Self Authenticating:  Based on the validation of the robust hash which was embedded.  It is to be extracted again from the test image itself. 

ii. Independently Authenticating:  Receive validation based on an image independent authentication sequence. Measurement of performance in terms of, i. Probability of miss when there is a forgery attack ,Pm ii. Probability of false alarm when there is no forgery,Pf 

Methods for comparison:  Semifragile watermarking algorithms should not give

false alarm against these permissible alterations: Mild compression. Histogram equalization. Sharpening Low-pass filter within a support of 3X3. Median filtering with a support of 3X3. Additive Gaussian noise down to signal to noise ratio of 35dB  Salt-n-pepper noise up to 1%  Random bit errors in transmission and storage of the image in row format, with a 0.001 probability of bit errors     



The nonpermissible alterations are the following:  Image forgeries intended to remove, substitute, or insert objects in the scene  Image manipulations that modify the geometry of the objects such as rotation, flipping or image manipulations that change the appearance of object such as color, shape, shadow manipulation,etc  Changes in the scene background.  Cropping

Semifragile Watermarking Algorithms: 

The expressions are,

 Image at pixel location (x,y) as I(x,y). Pixels of particular block b as Ib(x,y). Block Discrete Cosine Transform coefficients are indicated by Cb(p,q).



Lin-Chang Algorithm:  It is conceived to tolerate JPEG-style compression of watermarked image .  It is based on two properties DCT coefficient quantization, Order invariance: DCT coefficient pairs remain unaltered after JPEG processing if not set equal. Coefficient invariance: Coefficient is quantized to an integer multiple of the step size, its value is not changed after JPEG compression with smaller step size.

. . . . . .

Blocks

DCT C1(p,q)

. . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . .

Blocks

DCT C2(p,q)

1 0 1 0 0 1

Placed in DCT coefficient of same block pair



Authentication Data:  Consists of ordinal relationship of three pairs of DCT coefficients chosen from 8X8 blocks pair-wise coupled according to a random mapping.  For coupled block the authentication information at the (p,q) coefficient, Ø(p,q)={ 1 0



C1(p,q)-C2(p,q)>=0 C1(p,q)-C2(p,q)<0

Insertion Method:  If bits to be inserted complies with least significant bit of carrier DCT coefficient divided by the quantization step size, then no change is made.  otherwise the coefficient is incremented by a one step size amount.  This procedure should be iterated few times before DCT coefficients are stabilized.



Extraction and verification method:  It is replica of the insertion scheme.  Both the authentication bits are regenerated and LSB of the modulated DCT coefficients are read off.  An 8X8 block is declared as nontampered if 5 of 6 inserted bits are verified.  Otherwise block is tampered.



Lin-Podilchuk-Delp Algorithm:  Authentication sequence:  It is a pseudo-random zero-mean, unit-variance Gaussian noise sequence.  Its seed is controlled by a key but is otherwise independent of the document  Insertion Method:  The Gaussian authentication Sequence is placed in the upper triangular position of an empty DCT matrix  Then the inverse DCT of the matrix is calculated and resulting 16X16 spatial pattern is mixed additively with the image DCT block at a given strength. I’b(p,q)=Ib(p,q)+¥Wb(p,q) Finally block wise inverse DCT yields Semifragile watermarking image

 Extraction and verification method:  The watermark in the image is estimated by suppressing image spectral components in every block while enhancing the presence of the watermark.  Horizontal and vertical differences are calculated both for test image and spatial watermark pattern.  These difference vectors are concatenated to form two sets one derived from test image, I*b={ Dcol[Ib(x,y)] | Drow[Ib(x,y)] } other from watermark spatial image, W*b={ Dcol[Wb(x,y)] | Drow[Wb(x,y)] }



The verification is based on correlation of the original data with differenced version of the original watermark pattern. < I*b,W*b> Þ= [ <W*b,W*b> ]^1/2

Performance analysis of Lin-Changs Semifragile watermarking. 

Performance

watermarking.

of the Lin-Podilchuk semifragile

Experimental Results 

The following comments can be made on the integrated performance of Semifragile algorithms, on 64X64 blocks.

 Lin-Chang’s algorithm: The algorithm performs very well in the presence of JPEG compression, otherwise very fragile against signal processing attacks.

Lin-Podilchuk-Delp’s algorithm: It is robust against nonmalicious signal-processing operations, With Wiener filtering,to enhance the watermark signal, slightly improves the performance

Conclusion          

Semifragile watermarking. Alteration on documents. Types of water marking algorithms. Classification of tamper detection. Types of proposed algorithms. Measurement of the performance. Permissible and non-permissible alterations. Watermarking algorithms. Performace analysis of the studied algoritms. Experimental results.

September 21, 2009

JNNCE,Shivamogga Dept. of ISE

21





Special thanks to all staffs of ISE Dept. To my friends.