I'm doing research on image and audio watermarking, and so far I've read some literature in order to get basic introduction and good initial understanding of this topic. Right now I'm reading book "Digital Watermarking and Steganography" by Cox, et al.

My basic goal is to get know with robust watermarking techniques, especially for medical image marking.

If we look at the types of algorithms, some of the most commonly used (and robust) are Cox, Corvi and Xia algorithm, but all of them rely on watermarks generated as Gauss series of N pseudo-random numbers.

My question is, what if we want to embed custom watermark, like patient information (patient name, date of scan/record, etc.), or to use custom watermark as proof of ownership? Which (robust) algorithms and approaches are used in that case?

Thanks in advance.

  • $\begingroup$ what sorta signal would you be embedding these custom watermarks on? like an ECG signal? $\endgroup$ – robert bristow-johnson Jun 25 at 8:50
  • $\begingroup$ for now I would just focus on images like x-ray scans, CTs, etc, but also general RGB images custom watermarking.. $\endgroup$ – hideburn Jun 25 at 8:59
  • $\begingroup$ would these signals be degraded significantly if the LSB was stripped offa each sample word in the signal? in audio, we would call that a 6 dB loss of dynamic range. would a 6 dB loss of signal-to-noise ratio kill you? is your data that you want watermarked, is that floating point or fixed point? do you have bits to spare in the sample words? $\endgroup$ – robert bristow-johnson Jun 25 at 9:18
  • $\begingroup$ also, regarding steganography, do you want these watermarks to hide from people looking for them? with the intent to remove them upon discovery? that might be a little more difficult. $\endgroup$ – robert bristow-johnson Jun 25 at 9:22
  • $\begingroup$ yes, those signals must rely on steganography, for images they should be invisible and for audio they are on low freq unhearable range... I think LSB is a poor approach for this needs, because of huge degradation percentage under different types of attacks (i've tested this). For audio I'm not really sure if 6 db loss of DR would be a bad thing. $\endgroup$ – hideburn Jun 25 at 9:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.