Re:Don't dismiss watermarking too fast... :-(
on
SDMI *NOT* Cracked!?
·
· Score: 3
Let me try to explain why watermarks will fail: JPEG and MP3 are both streaming formats.
Consider an ID3 tag on an MP3 file. An ID3 tag is 128 plaintext bytes of data added to the end of an MP3 file so you can fill in the song's title, artist, album, etc. The ID3 tag is totally separate from the audio itself, since it comes at the end, and it does not change the audio one bit. This is very similar to the way GIF89a images can have text inside them; text which is totally separate from the image data because it comes separately at the beginning or end of the image data.
In stark contrast, a SDMI watermark is NOT separate from the audio itself. It is actual changes to the audio file itself interleaved throughout the audio data from beginning, to middle, to end. It is a series of actual frequency changes IN the audio, not outside of it at the beginning or end. It's a series of very slight changes, designed to be so small that they're inaudible but can be picked up by a watermark detector. The bottom line is that SDMI watermark DOES alter audio since it's PART OF the audio data and is not separate from it.
Also above you provided a link to a site about JPEG watermarking and the claim is that the watermark does not affect the JPEG image. JPEG and MP3 are very close relatives. JPEG watermarks DO affect the image data, ever so slightly, if it's interleaved. You can't see it because it might be a difference of 1 or 2 shades per pixel on a palette of 16777216 colors (JPEGs are 24-bit by default). a JPG with a non-altering plaintext watermark inside it would be invalid and unviewable, since you'd have image data and non-image data together. Like, the image would cut off where the watermark is, and start off further down out of alignment. It's the same concept behind skips in MP3s. If you miss 10kb or so due to a bad download, you'll be able to hear the rest of the song but there are frames missing and there is a loud glitch. So SDMI is a data-altering watermark which alters the audio and is not the audio the artist intended his audience to hear. It's amazingly close, but it's not the same.
Anyway, i don't know yet about 100% eradication of music watermarks, but when it comes to killing image watermarks, this works great for me: Just open the watermarked image in your favorite photo editor, highlight the entire image from very top left to very bottom right, and copy&paste that as a new image. Don't just copy&paste the whole image.. actually highlight from top left to bottom right and paste as a new image, new selection. There goes the watermark! Poof.. gone.
http://www.cyberdeck.org/countzero/techa.html is very good for analysis and i appreciate the reply so fast. but a "spectrogram" is not a way of zooming in and looking at a tiny sample of the wave like that website does, it's a view of the entire wave with elapsed time on the X axis and frequency response on the Y axis and is far more revealing about things like watermarks and inaudible data added to sound. with a spectrogram we could make many more educated assumptions about these watermarks, the formula behind them if any, etc. without doing wav>mp3>wav or digital>analog>digital conversions and merely hoping for the best, so if someone has a spectrogram of the original HackSDMI wav files, or if someone has the actual wav files where i can download them, please let me know either way. thanks.
sorry if this has been asked before, but does someone actually have the original 6 files that were posted on http://hacksdmi.org that were supposed to be the challenge?? i'd like to view the spectrogram of those files to see if there is a visible watermark in the frequency distribution. i've heard that these watermarks are supposed to be able to survive the conversion from digital to analog and back again, so i don't see how that can be true unless they're visible (through a spectral view of the waveform) or clearly audible through anyone's ears.
thanks. d;d
Slashdot Mirror
Let me try to explain why watermarks will fail: JPEG and MP3 are both streaming formats. Consider an ID3 tag on an MP3 file. An ID3 tag is 128 plaintext bytes of data added to the end of an MP3 file so you can fill in the song's title, artist, album, etc. The ID3 tag is totally separate from the audio itself, since it comes at the end, and it does not change the audio one bit. This is very similar to the way GIF89a images can have text inside them; text which is totally separate from the image data because it comes separately at the beginning or end of the image data. In stark contrast, a SDMI watermark is NOT separate from the audio itself. It is actual changes to the audio file itself interleaved throughout the audio data from beginning, to middle, to end. It is a series of actual frequency changes IN the audio, not outside of it at the beginning or end. It's a series of very slight changes, designed to be so small that they're inaudible but can be picked up by a watermark detector. The bottom line is that SDMI watermark DOES alter audio since it's PART OF the audio data and is not separate from it. Also above you provided a link to a site about JPEG watermarking and the claim is that the watermark does not affect the JPEG image. JPEG and MP3 are very close relatives. JPEG watermarks DO affect the image data, ever so slightly, if it's interleaved. You can't see it because it might be a difference of 1 or 2 shades per pixel on a palette of 16777216 colors (JPEGs are 24-bit by default). a JPG with a non-altering plaintext watermark inside it would be invalid and unviewable, since you'd have image data and non-image data together. Like, the image would cut off where the watermark is, and start off further down out of alignment. It's the same concept behind skips in MP3s. If you miss 10kb or so due to a bad download, you'll be able to hear the rest of the song but there are frames missing and there is a loud glitch. So SDMI is a data-altering watermark which alters the audio and is not the audio the artist intended his audience to hear. It's amazingly close, but it's not the same. Anyway, i don't know yet about 100% eradication of music watermarks, but when it comes to killing image watermarks, this works great for me: Just open the watermarked image in your favorite photo editor, highlight the entire image from very top left to very bottom right, and copy&paste that as a new image. Don't just copy&paste the whole image .. actually highlight from top left to bottom right and paste as a new image, new selection. There goes the watermark! Poof.. gone.
http://www.cyberdeck.org/countzero/techa.html is very good for analysis and i appreciate the reply so fast. but a "spectrogram" is not a way of zooming in and looking at a tiny sample of the wave like that website does, it's a view of the entire wave with elapsed time on the X axis and frequency response on the Y axis and is far more revealing about things like watermarks and inaudible data added to sound. with a spectrogram we could make many more educated assumptions about these watermarks, the formula behind them if any, etc. without doing wav>mp3>wav or digital>analog>digital conversions and merely hoping for the best, so if someone has a spectrogram of the original HackSDMI wav files, or if someone has the actual wav files where i can download them, please let me know either way. thanks.
sorry if this has been asked before, but does someone actually have the original 6 files that were posted on http://hacksdmi.org that were supposed to be the challenge?? i'd like to view the spectrogram of those files to see if there is a visible watermark in the frequency distribution. i've heard that these watermarks are supposed to be able to survive the conversion from digital to analog and back again, so i don't see how that can be true unless they're visible (through a spectral view of the waveform) or clearly audible through anyone's ears. thanks. d;d