Hacker News Reader

Comment by rockyleal | Hacker News Reader

"media files often contain lots of entropy that could help identify the author, such as location information"

Please, please, explain this.

Hidden image data in photos that I know about:

(1) Exif meta-data (Exchangeable image file format) which are things like resolution and dates, but can include serial numbers as well. Here's a handy list of cameras (usually high-end ones) that betray your privacy by embedding a serial number:

http://www.stolencamerafinder.com/listmodels

(2) IPTC meta-data (International Press Telecommunications Council), though I have yet to see a photo "in the wild" that uses it.

(3) XMP meta-data (Adobe Extensible Metadata Platform). This is added by Photoshop for example. It's two long numbers (DocumentID and InstanceID) which are different for each image.

(4) DRM and anti-counterfeiting stuff that might be added by image editing tools. (Many color copiers do this in the case of paper prints, for example.)

I want to add some tips:

Exif is the one everyone knows, but don't discount #3 and #4 (i.e., identifiers added by image editing tools).

To keep a photo untraceable:

Delete the Exif data using a tool like IrfanView.

To edit the photo, download a fresh copy of an open source editing program or a commercial editing tool that does not need registration or activation. After you're done working on the photo, delete the editing program.

For a higher level of security (to guard against the possibility of a GUID or hostname from the OS being embedded in the image), do all of the above on a computer that is not traceable to you, or on a fresh install of the OS, which you wipe after you are finished.

Also, seed the network card with a fresh random MAC on eevery reboot, or more often.

"betray your privacy" is a pretty loaded way to say that.

Sensor noise can also be used to identify pictures taken with a specific camera:

http://wrap.warwick.ac.uk/3318/

If I were being truly paranoid, I'd use some steganography software to write random "data" across all the low-order bits of any images I _really_ didn't want traced back the the sensor in my camera/phone/videocamera...

As an example, take a photo with a relatively new digital camera and open the JPG up with an EXIF reader. In all likelihood, you'll see GPS coordinates, the exact time you took the picture, along with the unique identifier of your camera.

See http://en.wikipedia.org/wiki/Exchangeable_image_file_format#....

That's surely not what kijin was referring to; that can easily be stripped out. Rather there's the problem of being able to determine location purely from the picture itself -- a picture contains a lot more information than just the particular thing you intend to point out in it.

Even if you scrub the EXIF data (which you absolutely should do of course) remember that it's likely possible to identify if you took a particular photograph with a particular device.

That isn't a danger that you'll be identified from the photo, but it prevents plausible denial at a later date.

https://www.schneier.com/blog/archives/2006/04/digital_camer...

This can be easily overcome by adding random noise.

Random noise only means you need more samples; it can't block a consistent pattern.

You can add a consistent random pattern.

You need a method that can't be distinguished from the native pattern. How much trust do you want to put in your algorithm being invulnerable to future analysis?

blur and threshold.

Quick and dirty:

  cat imagefile | strings | less

To obfuscate meta-data in media files, use two different editors to covert file formats three times. eg Use Picasa to convert original to high quality jpeg, then use Gimp to convert jpeg to png

Reply on news.ycombinator.com