基于高斯混合模型的多尺度的脆弱的数字水印

Multiscale Fragile Watermarking Based on the Gaussian Mixture Model
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 15, NO. 10, OCTOBER 2006

 

Thispaper is mainly in the description:

1. Images of wavelet transform(DWT), re-use of probability density function (PDF) The concept ofWatermarking Block to calculate the value of each V, Eventually, by theCode Map to the watermark into embedding in the image. For example, if my CodeMap is designed to embedding "00", one that I passed the upper rightcorner after the DWT transformation Watermarking Block V-to be equal tothe lower left corner of the V, so I only have to modify one of them Vso that they are equal to each other, but to remove the watermark, whenthey see top right, bottom left corner of the V are equal, you can knowthat the watermark Watermarking Block possession and for "00."
2. It have mentioned that the size of each Watermarking Block can not be lessthan 16 * 16, which is too small to make a Block size of ourV-representation of lower demand, on the contrary, if the Block sizelarger, we obtained The stronger correlation of V, that is if the Blockhas any one pixel has been changed, the easier it is sensitive to thevalue reflected in V above.  While another advantage is that when we are hiddenwatermark into the time to let the two different Block V-equivalent,and when Block size larger, we modify the value of V can be more evenlydistributed and then the top of the pixel , and thus the destruction ofthe original image is also less, so this method in the possession ofthe author into the watermark, the image PSNR can be as high as 50 dBor more.
3. The shortcomings of paper here, I think is his ability to watermark the amount of possession into the bar. A Watermarking Block can not be less than 16 * 16, and each Block can only be hidden 2 Bit information.  If a 512 * 512 images, finished third order of the DWT transformation, can only be hidden into the 680 Bit. But I think he has a watermark I have read the relevant papers and inthe contribution of people do not like is that he can be a very minordamage to very sensitive detection, for this reason there is a simplereference to the above, he is to go with a Block calculated a V value,and this Block inside the pixel has been modified, it will underminethe relevance of the whole play, leading to V values before and afterthe not the case. And then his results, there is the inclusion of a 0.0001% of theGaussian noise, the author's method can easily detect the image to bedestroyed. In addition, he can also distinguish this method Malicious (eg. crop, blur... etc.) and non-malicious (eg. compression, brightness adjustment)difference between the two.
4. The main method is to watermark hidden in the multi-step of the DWT,the malicious destruction of when the order of destruction willsignificantly increase the higher the contrary, the non-maliciousdamage, the order of increasing extent of the damage coal is almost nochanges.

Experience:

 

        The paper is proposed to watermark DWT to embeding, which is alreadywell-versed in the technology, but coupled with his chances ofreunification with the machine method and the block DWT after there-converted into Watermarking Block, to amend the Block to mention awhole values, so that each Block possession of the watermark into the 2Bit.Can then be very sensitive to detect a watermark image in possession ofthe damage, and can distinguish between damage of malicious ornon-malicious.

        Indeed, in this regard is indeed greatly enhance the strength and toughness of the watermark. Although it can not be more than the amount of possession, but thespirit of the original watermark is to protect copyright, but not in the possession of the amount of income.

        Therefore, a strong non-destructive watermarks, is what we want!