Spatial Domain Digital Watermarking of Multimedia Objects for Buyer Authentication By: Ravi Andem Todd Yarrington (randem@mix.wvu.edu) (tyarring@mix.wvu.edu)
Motivation Encryption of valuable images Keep track of original digital images being distributed to buyers Copyright protection
Main Idea The work on spatial domain digital watermarking of multimedia objects for buyer authentication by Mukherjee, Maitra and Acton exhibits the following: Image owner authentication Survival of watermark Spatial domain Frequency domain No visual degradation
Overview of Presentation 1. Introduction of Work 2. Generation of Image and Buyer Keys 3. Watermarking 4. Analysis of Results 5. Limitations 6. Suggested Improvements & Future Work 7. Conclusion
1. Introduction of work Use watermarking to encrypt images What is watermarking? Basically a hidden message Text Image Data, etc. In this case, the authors use a unique binary vector to decide where to hide numbers in certain groups of pixels
1. Introduction of work (cont d) Watermarking Process
2. Generation of Image and Buyer Key Image Key Random spatial division of the multimedia object into a set of disjoint subsets Part of image to be watermarked 2 2n-1 (a+b-1) = possible image keys/image Image key of the same part of image
2. Generation of Image and Buyer Key (cont d) Buyer Key Binary vector, B Length of B = m = 2 n = # of groups B β B is selected from C C is set of error correcting codes M distinct codewords in C (with Hamming Distance at least d) β is the number added to the intensity at a position
2. Generation of Image and Buyer Key (cont d) Reed-Muller Code Set of 2 n bit long codewords (CW) Containing M=2 n+1 distinct CWs with minimum distance d=2 n-1 Example: for n = 2, d = 2 0000 0011 0101 1010 1100 0110 1111 1001
3. Watermarking Insertion of Watermark Original image is spatially divided into a number of blocks based on image key Image intensity of each block is modulated depending on bit values of buyer key Group 0 1 2 3 Example: β = 1 0-1 0
3. Watermarking (cont d) Retrieval of Watermark Original and Watermarked images compared block by block (block info. obtained from the image key) Depending on extent of intensity modification in each block, a probable buyer key is generated Generated buyer key is mapped to exact buyer key by correcting the errors using correcting codes
3. Watermarking (cont d) Identifying Buyer Key From Attacked Watermarked Image Image information may be changed so much, σ k, that it could be interpreted by the retrieval system as something completely different Toleration factor c k is used for weighting δ k
3. Watermarking (cont d) Analysis of Watermark Retrieval Process
3. Watermarking (cont d) Resistance to Nonlinear Attacks Nonlinear attacks include: JPEG Compression attack Stirmark Geometric attack
Implementation 256X256 Peppers Input Image Divided into 128 groups (512 pixels per group) 128 bit Reed Muller Code for generating buyer key Simulated Spatial and frequency domain attacks
4. Analysis of Results Cryptographic Robustness in Spatial Domain Results Robust to spatial domain attacks such as: Low pass filtering Median filtering Rotation Robust to frequency domain attacks
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Low-pass filtering
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Low-pass filtering
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Median filtering
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Median filtering
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Rotation
4. Analysis of Results (cont d) Performance Against Spatial Domain Attack Rotation
4. Analysis of Results (cont d) Performance Against Frequency Domain Attack
4. Analysis of Results (cont d) Performance Against Frequency Domain Attack
4. Analysis of Results (cont d) Authentication of Buyer Key Find the codeword q closest to q q is the retrieved buyer key Incorrect retrieval is improbable over the complete range of possible tolerance factors
5. Limitations Limitations of the paper include: If number of errors exceeds d/2 then the scheme fails Does not use a buyer specific image key
6. Suggested Improvements & Future Work Use non-uniform values of β(i,j). This can increase robustness to attacks on watermarked image Use a buyer specific image key Simulate nonlinear attacks like JPEG compression.
7. Conclusion Introduced to the problem of Identification of Copyright material Proposed method survives attacks in: frequency domain spatial domains Proposed Improvements
Thank you! Any questions/comments/suggestions?
References D.P. Mukherjee, S. Maitra, and S.T. Acton. Spatial Domain Digital Watermarking of Multimedia Objects for Buyer Authentication. IEEE Trans. On Multimedia, Vol. 6, No.1, Feb. 2004.