EE4414 Multimedia Communication System II, Fall 2005, Yao Wang Homework 10 Solution
Download 163.49 Kb.
|
| EE4414 Multimedia Communication System II, Fall 2005, Yao Wang
Homework 10 Solution (Data Hiding, Watermarking, Steganography) ___________________________________________________________________________________ Review questions Describe the differences between data hiding, watermarking, and steganography. What are their respective applications and requirements? Is one a special case of another? Data hiding refers to a technique that embeds data in a cover media. Watermarking is a special type of data hiding, where the embedded data carries information that can be used to prove the ownership or integrity of the cover media. Steganography is also another special type of data hiding, wherein the embedded data is a secrete message that is typically unrelated to the cover media. The cover media can be chosen randomly or to facilitate the conveyance of the secrete message. Because multimedia documents (audio or image or video) have a lot of redundancy, these documents are often chosen as the cover media for embedding secret messages (which themselves could be an image or plain text or encrypted text). With watermark, the embedding algorithm should be designed so that the embedded watermark is hard to remove (except for applications using fragile watermark). On the other hand, for steganography, this is usually not a concern. The following are some possible applications of watermarking. Describe what they are and their requirements. Ownership assertion Fingerprinting Copy prevention and control Detection of unauthorized alteration (authenticate integrity of the data) Ownership assertion: embedded watermark is used to proof the ownership of the cover media. The owner or its trusted delegates should be able to detect the presence of the watermark (with the help of the original and the secret key used to generate the watermark) even if the watermarked document is subjected to accidental or malicious alterations (such as compress/decompress, cropping, printing/scanning, shrink and then zoom). Fingerprinting: a different watermark is embedded to each sold copy of the original media. If this copy is altered or being reproduced illegally, based on the embedded watermark, the owner can track down the person who purchased the copy. Copy prevention and control: the watermark contains information on how many copies can be made. A compliant recorder/copier will detect such watermark before making copies. Detection of unauthorized alteration: the watermark should be such that if the media with the watermark is altered, even if slightly, then the extracted watermark is different from the original watermark. What is the difference between robust and fragile watermarks? Which technique you will use for the above applications (a-e in Prob. 2)? Robust watermarks are those that can withstand accidental or malicious manipulations of the watermarked document. Fragile watermarks are those that will change if the watermarked document is somewhat manipulated. Robust watermark is useful for ownership assertion, fingerprinting, and copy control, whereas fragile watermark is used for prevention/detection of illegal alteration of a document. Propose one possible technique, for embedding a watermark (invisible) in an image for ownership assertion. Also describe the corresponding watermark detection scheme. Remember that the watermarked image should look almost identical to the original image, and that the watermark should be detectable even after the watermarked image is modified somewhat (saved with a different compression ratio; printed, copied and then scanned; etc.) See slides 29 and 30 in lecture note “memon_F05.pdf” for two possible techniques: one adds a pseudo random sequence in image sample domain directly, another in DCT coefficient domain. More specifically, in the transform domain technique described on Slide 20, the image is subject to block-wise DCT transform, then all (or selected blocks based on some secrete keys) DCT coefficients with intermediate frequencies (say those corresponding to row 3-4, and column 3-4 in 8x8 DCT) are modified by adding some random sequences (watermark signal). For example, if the chosen DCT coefficients are arranged into a 1D sequence, {f_1,f_2,…,f_N}, and watermark signal is represented by {w_1,w_2, …,w_N}, where w_i are zero-mean random variables with small magnitudes (e.g., only –1,1 are possible values), these DCT coefficients are changed to {f_1+w_1, f_2+w_2, …, f_N+w_N}. The image with changed DCT coefficients are then inverse-transformed back to the image domain. Because w_I have small magnitudes, they should not induce visible changes in the watermarked image. Only the mid-frequency DCT coefficients are modified, to satisfy the perceptual similarity and robustness requirements. If the very low frequency DCT coefficients are modified, the watermarked image is likely to have visible artifacts. On the other hand, if very high frequency coefficients are modified, the added watermark can be easily removed by high-pass filtering. To detect the watermark, one can do the block-wise DCT on the watermarked image, select the same set of DCT coefficients,  and correlate these coefficients with the watermark signal (the owner knows which coefficients are modified and the watermark signal), by calculating  . If the watermarked image is not altered at all, then  Even if the watermarked image has been subjected to some manipulations, it is likely that C is a large number greater than T=N/2. Therefore, as long as C>=T, one can verify the presence of the watermark.
Download 163.49 Kb. Share with your friends:
|