Reversible watermarking

Wu Dan2008.2.20

Introduction

What?

Introduction

Why? Military data Medical data

How? Data compression

Difference expansion Histogram bin shifting

Reversible Data Embedding using a Difference Expansion

Jun Tian

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL.13 NO.8 AUGUST 2003

How to measure a reversible data embedding algorithm?

Payload capacity (bpp) Visual quality (PSNR) Complexity

A simple example of the difference expansion

x=206, y=201; b=1.l: the integer averageh: difference

DE: difference expansion

The new values:

Reversible data embedding

Reversible integer transform

The inverse transform:

To prevent the overflow and underflow :

Expandable and changeable difference values Expandable:(for both b=1,0)

Changeable:(for both b=0,1)

By definition, we can find that: If h is changeable, h’ is still

changeable. If h is expandable, h is changeable. After the DE, the expanded

difference value h’ is changeable. if h=0 or -1, the conditions on

expandable and changeable are equivalent.

Data embedding algorithm :1. The original image is grouped into

pairs of pixels values. Then compute the difference values h.

2. Create four disjoint sets of difference values: EZ, EN, CN, and NC

EZ: contains all expandable h=0 and

expandable h=-1. EN: contains all expandable h

EZ

CN: contains all changeable

NC: contains all non-changeable h.

3. Create a location map of selected expandable difference values.

4. Collect original LSBs of difference values in EN2 and CN. However for those h=1 or h=-2 in EN2 and CN, their LSBs will be not collected.

5. The location map will be losslessly compressed. The compressed bit stream is denoted as L. Embed L, the original LSBs C, and a payload P.

6. Apply the inverse integer transform to obtain the embedded image.

Discussions: Capacity:

Threshold:

The scanning order: Non-changeable:

Scanning order :

Non-changeable:

decoding :1. Calculate the difference values h.2. Create two disjoint sets of difference val

ues: CH and NC changeable and non-changeable3. Collect LSBs of all difference values in C

H, and form a binary bit stream B.4. Decode the location map from B, and res

tore the original values of differences as follows:

Experimental results:

Alattar

Jun Tian

Chin-chen Chang

Dinu Coltuc

Reversible data hiding

Zhicheng Ni, Yun-Qing,Nirwan Ansari, and Wei Su

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY,

March 2006

Algorithm

Zero point and peak point

Embedding: Generate the histogram H(x). In the histogram, find the zero point H(a)

and peak point H(b). If H(b)>0,record the coordinate of those

pixels. Assume a<b. Scan the image. If x∈(a,b),x+1; leaving the value a+1 em

pty. If w=0, a=a; if w=1,a=a+1.

Pure payload: C=H(a) - H(b) Multiple pairs of Maximum and minim

um points:

Extraction algorithm: ( Assume the zero point and peak points are a ,b ) Scan the image in the same order as in the e

mbedding procedure. If the value is a+1,w=1; if the value is a, w=0. Scan the image again, if the grayscale value

x∈(a,b], x-1. If the overhead information found in the extr

acted data, set the pixel grayscale value as b.

Lower bound of the PSNR of a Marked image

The total embedding time is just 100ms.

Experimental results

Discussion: 1) How to get the peak point and

zero point for verifier? 2) How to use the a and a+1?

Reversible watermark using the difference expansion of a generalized integer transform 　

Adnan M.Alattar, Member, IEEE,

IEEE TRANSACTIONS ON IMAGE PROCESSING, AUGUST 2004

Generalized difference expansion

Vector:

Reversible integer transform:

return

A difference expansion oriented data hiding scheme for restoring the original host images 　

Chin-Chen Chang, Tzu-Chuen Lu

The Journal of systems and software,May 2006

return

Very Fat Watermarking by Reversible Contrast Mapping 　

Dinu Coltuc and Jean-Marc Chassery

IEE SIGNAL PROCESSING LETTERS,APRIL 2007

Reversible contrast mapping:

Dc: the domain without the odd pixels pairs. Embedding: 1 partition the entire image into pairs. 2 for each pair: a) if (x,y) is even pixel pair, set the LSB x’ t

o 1, the LSB of y’ is the watermark. b) if (x,y) ∈Dc, set the LSB of x to 0, and the

LSB of y is the watermark.

c) if (x,y) Dc, set the LSB of x to 0, and save the ture value.

return