狂贺建民大联盟首胜.pptVIP

Rotation, Scale, and Translation Resilient Watermarking for Images Ching-Yung Lin, Min Wu,Jeffery A. Bloom, Ingemar J. Cox, Matt L. Miller, and Yui Man Lui Introduction(1/2) The issue which we are concerned with:- Robust to RST distortion- Blind watermarking Our Proposal:- Based on developing a watermark that is invariant to geometric distortions, thus we can eliminating the need to identify and invert RST Introduction(2/2) Our Strategy:- Transfer spatial domain to a domain which is invariant to RST- Embed watermark in this domain - Transfer to spatial domain Important Problem:- What is the domain which is invariant to RST? And how to transfer?- How to embed watermark? Or how to design watermark? Algorithm – Invariant Property (1/4) Consider an image i(x,y) and a RST version of the image i’(x.y): The Fourier Transform is I’(fx , fy):Cool! It is independent of the translation parameters!But how about rotating and scaling? Algorithm – Invariant Property (2/4) We try to transform it into log-polar coordinates: Then the magnitude of I’(x,y) can be rewritten as: or Algorithm – Invariant Property (3/4) See above equation we find:- the image scaling result in a translation shift - the image result in a cyclical shiftp.s we need not concerned with the amplitude scaling of the spectrum, since we perform watermark detection using the correlation coefficient. Algorithm – Invariant Property (4/4) For find more invariant value, we define: It would be more better to compute the value below:- It is invariant to both translation and scaling. Rotation result in a shift of its values. Algorithm – WM Detection Process (1/3) The watermark w is expressed as a vector of length N. A extracted signal v from the test image is compared with w by using correlation coefficient. Step:- Compute a log-polar mapping table as below from input image. Algorithm – WM Detection Process (2/3) - sum all the values per column j and column j+n/2. Then, we sum t