Image steganography and tracking

Question

We are running an online-press media-library. On this site many brands are uploading images and track press users download and views per image. But some brand now want to add unique code in image so they can track their own images.

We tried some steganography tools:

1. http://www.outguess.org/info.php (support JPG)
2. http://www.imagemagick.org/Usage/transform/#stegano (support PNG)

Both works fine, but issue is that secure code disappeared after image resize or crop.

Simple way would be watermark, but we don't want to disturb images, because images are being used in press printing.

Is there any way to put secure code in JPG format image and track that code after resize or cropping of that image?

Update

Our brands want to publish them as web image(light version), so press users could use images on blog and forums.

We will target press blog / forum site from crawler engine(~4500 blogs) and check if our image exist on these blogs.

Answer 1

You can add information to JPEG images in two ways:

• you add information out of any image data, that is to say the JPEG headers
• you add information in the image data, that is to say you modify the image

The first technique can be easily bypassed by simply removing the relevant data from the header.

The second technique can be broken down into:

• Watermarking
• Steganography

Steganography is used when you want to distribute information in images and want to remain as stealth as possible. You don't say you are putting information into the images and you try by all mean to minimise your impact on the image statistics.

Watermarking is used mostly for DRM purposes, you want to put information in the image and ensure it stays when modifications are made to this image (it is called a robust watermark). Robustness usually comes at the price of image alteration (we can say degradation here), there is no actual way of achieving a robust watermarking without "disturbing" the image as you say. There are many, many techniques of watermarking (which are off-topic for this question) with different type of robustness (cropping and resizing included) with different impact on quality. You should ask google scholar about that.

You'd probably better have two copies of the image, one for online display (with watermark) and one for printing (without watermark).

Edit: since you're not convinced yet, here is some link to relevant articles:

• A Comprehensive Survey of Contemporary Researches in Watermarking for Copyright Protection of Digital Images
• Geometric attacks on image watermarking systems
• A Survey of Digital Watermarking Scheme
• Semi fragile watermark with self authentication and self recovery

they should give you a good overview of the state of watermarking and its properties and possible attacks on it.

For further inquiry see at : https://scholar.google.com/

Answer 2

You might get good mileage out of merely adding this information as JPEG EXIF data, which will survive most modern image manipulations ... but is easily stripped out by somebody wanting to remove it. (A downside: PNG, GIF, and JPEG 2000 do not support EXIF. See my update below for XMP, which doesn't have that issue.)

Just to reiterate, EXIF data is rather well organized, easily viewed, and easy to remove (or modify), so it is not a security solution, but I don't think you'll find a viable security solution without watermarking the image.

If you don't want people to see the contents, encrypt it or use a code the requires looking up in your database. Then all somebody can do is remove it or modify it (which will destroy it).

Of course, if a database is on the table, you might be able to save a 32x32 pixel grayscale version of the image and use that as a hash (you'd be surprised at how few collisions you get). This will survive resizing and minor color correction, but it will not survive aggressive cropping or other manipulations.

EXIF and the db are not mutually exclusive; you could do both.

---

Update: I wrote this about EXIF, which is decently well understood and supported, but only supports JPEG. You could consider XMP, which supports pretty much all image types (and beyond, e.g. PDF and MP3) instead. I'm not as sure about its ability to survive edits made by editors that may not support it, but it looks pretty supported.

The XMP Wikipedia article doesn't mention ImageMagick support, and a quick web search reveals that ImageMagick probably supports XMP data; you'll want to verify that before deploying. I'd advise against XMP if edits from ImageMagick end up removing it. (ImageMagick is extremely common in web apps and shell scripts.)

Answer 3

Your're probably dealing with two different causes for loosing the hidden data:

• Resize: this will operation will use an image scaling algorithm to create a new scaled version of the image, all pixel data (including your hidden data) will be lost.
• Crop: some of the hidden data may be stored in the areas removed by the crop, then typically upon saving the cropped image a fresh pass of jpeg compression will be applied which will also destroy the hidden data.

I'm pretty sure that image owners such as Getty use complicated algorithms to decide if a given image is a match to an image in their library. Without the use of watermarking (something visible on the image) I can't think of another option I'm afraid (well none which wouldn't be easy to remove anyway).

Hope this helps!