Digital Watermarking SIMG 786 Advanced Digital Image Processing Mahdi Nezamabadi, Chengmeng Liu, Michael Su
Different types of Digital Image Watermarking Visible: watermark is a secondary translucent overlaid into primary image. Invisible-fragile: invisible, any modification of the image will destroy the watermark. Invisible robust: watermark is perceptually not noticed and it can be recovered only with appropriate decoding mechanism. It’s robust to friendly or malicious attacks.
Image Watermarking Using Phase Dispersion: Embedding I’(x,y) Watermarked image I(x,y) Source image (no watermark) M(x,y) Message Image, to be embedded, preferable to use the edge maps of an icon
Image Watermarking Using Phase Dispersion: Embedding C(x,y) Carrier function. It is generated by the private key. It has random Fourier phase and non uniform magnitude An arbitrary constant α chosen to make the embedded message simultaneously invisible and robust to common processing Tiling the original image and embed the same image in each tile independently improves the robustness
Embedding process illustration
Image Watermarking Using Phase Dispersion: Extraction Extraction function M(x,y) can be calculated from M’(x,y) For a carrier with uniform amplitude
Image Watermarking Using Phase Dispersion: Extraction
Carrier Function Design Consideration P(x,y) denotes the autocorrelation function of the carrier function In order to improve the extracted image quality, it should be as close to a delta function as possible Human visual system falls off rapidly with increasing spatial frequency
Carrier Function Design Most of the carrier energy should be concentrated in high frequencies to make invisible The phase of the carrier is generated using pseudo-random number generator with a user-specified key The magnitude is set to 0 at 0 frequency (DC value)
Carrier Function Design Magnitude gradually increased with increasing spatial frequency up to about 1/5 of Nyquist frequency For frequencies greater than 1/5 of Nuquist frequency, the carrier envelope is derived from the Contrast Sensitivity Function (CSF) data The CSF provides a measure of sensitivity of the average observer to changes in contrast at a given spatial frequency
Contrast Sensitivity Function Reciprocal of the CSF can be used to determine the carrier magnitude needed at a given frequency to bring the embedded signal just below the threshold of detectability by an average observer
Message Template Design T(x,y): Message Template Function, the image resulting from placing a positive delta function at every message location This is for binary message
Rotation/Scale Detection and Correction Moment normalization, set local mean of the watermarked image to 0 and its standard deviation to a target value σ d Do autocorrelation on the processed image and then process with a high-pass filter
Rotation/Scale Detection and Correction The ability to handle rotation and scale is a fundamental requirement of robust data embedded techniques
Thank you, Questions?