Refer to the following paper:

Ingemar J. Cox, Joe Kilian, F. Thomson Leighton and Talal Shamoon, "Secure Spread Spectrum Watermarking for Multimedia," IEEE Trans. on Image Processing, Dec. 1997

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=650120

We will test the method for two types of detection Non blind and Blind.

Case I: The Non blind Detection case

Follow the Embedding process as:

1) Generate a watermark W of 1000 bits consisting of {0,1} from Normal distribution and convert the watermark to {-1,1} why??.

You may like to use Signum function for this conversion.

2) Transform the cover image using NxN DCT and select 1000 largest coefficients except DC.

3) The watermark strength ß is taken as 0.1 and use additive rule for embedding.

Use the following detection process:

1)Use an original cover image for detection

2)Transform the image using NxN DCT.

3)Find the 1000 largest coeffcients in I' received cover.

4)Perform I'-I/ß to recover the watermark W'.

Considering that image has not been manipulated answer the following

1)Perform the normalised correlation(NC) W.W'/sqrt(W.W') and write your observation.

2)Create 10 more watermark patterns W''(do not embed them) and again perform Normalised correlation with recovered watermark W'.

3)Plot Normalised correlation against each watermarking pattern and write your observations.

Case II: Non blind detection

Use the following algorithm for generation and detection

1)Generate a watermark W of 1000 bits consisting of {0,1} from Normal distribution and convert the watermark to {-1,1}. You may like to use Signum function.

2) Transform the cover image using NxN DCT and select 1000 coefficients in Zig Zag Order except DC. (For Zig Zag Encoding Pl. refer to JPEG compression).

3) The watermark strength ß is taken as 0.1 and use additive rule for embedding.

Detection:

1)The received image is transformed using NxN DCT.

2)Find the first 1000 coefficients in Zig Zag fashion V'.

3)Detect the watermarking bits from them.

Considering that image has not been manipulated answer the following

Task: Peform the Normalise correlation between the recovered coefficients V' and Watermark W.

2)Create 10 more watermark patterns W''(do not embed them) and again perform Normalised correlation with recovered Coefficients V'. Plot Normalised correlation against the particular watermark pattern and write your observation.

Does the Normalize correlation value remains same as Non blind case???.

Experiments Instruction:

Use a gray scale image of size 512 x 512 in TIFF or uncompressed formats.

Subject the watermarked image to following attacks (for non blind detection) and compute the NC between recovered watermarking bits W’ and original watermark W for each attacks.

1)JPEG Compression with Quality factor Q=50.

2) Rotation by 100and Cropping Cropping.

3) Low pass filtering with an mask size of 3x3.

4)Reducing the image dimension to half.

Available from: Saturday, 31 January 2009, 5:30 PM
Due date: Monday, 16 February 2009, 3:30 PM