Principal component analysis (PCA) has found a broad range of applicability in questions of digital content protection and data hiding.  The field of steganography, in particular, relies upon PCA as a means of embedding one stream of data (for example, a watermark, or a copyright indicator) into another stream of data in a way that is undetectable in most circumstances, and yet robust enough to require severely distorting the watermarked file in order to remove the watermark.

With images, steganography aims to allow artists and photographers to ensure they have the ability to enforce claims of ownership and copyright without forcing them to “destroy” their images by placing large, visible, hard-to-remove visual watermarks on the visible portion of the image.  For bitmap images, steganography takes advantage of the inability of the human eye to detect extremely subtle variations in color.  For example, in such an image, each pixel is capable of representing 2⁸ different shades of blue.  The difference in color between, say, the shade of blue represented by 10011110 and 10011111 (e.g. a change in only the least significant bit) is likely to be undetectable by the eye, and can thus be used to encode a portion of the other data stream.  As each pixel has three components (R, G, and B, ignoring the A as this is a bitmap image), we can store one byte of non-color data for every three pixels in the source image.

As an example, consider the following image, with a second image embedded inside it. Note that because the embedded image is, in essence, just another stream of data, arbitrary files can be embedded in this source image and recovered using similar techniques.

If you use a computer program to remove all but the last two bits of every pixel in this image, an image that appears completely black will result.  If, however, you then take the resulting image and increase the brightness by 85 times, the following embedded image results:

Note that the requirement that this be a bitmap (e.g. uncompressed) image is crucial: the PCA transforms that image compression schemes use during quantization introduce distortions into these pixel values, often  cutting out the subtle, imperceptible variations in shades of color that we just exploited in order to reduce file size, thereby corrupting the embedded data.  However, steganographic techniques that rely on the specific operation of an image quantization algorithm like JPEG are available, and can be used to embed data into the compressed version of the image.

With music, steganography allows creators and distributors of compressed audio to embed ownership and copyright information into the quantized audio stream in a way that makes it virtually indistinguishable from an unaltered stream, yet can be extracted if necessary to prove ownership.  Here, steganography works with the quantized stream, modifying the encoder’s parameters to increase the available space for data storage pursuant to the psychoacoustic encoding model the compression scheme uses to cut out the portions of the frequency range that are inaudible to human ears.  Programs like MP3Stego allow arbitrary data to be introduced into compressed MP3 audio files, and offer a good measure of protection against removing the embedded data.  Because the embedded data becomes, in effect, part of the sound that is being encoded, it is impossible to remove the data without recompressing the file, at a severe and easily noticeable loss in quality.  This is the same reasoning behind watermarks in compressed digital images: the embedded data becomes part of the image data that is being compressed.  This property makes steganography quite promising in society today, as issues of copyright and content protection are undergoing close scrutiny.

References

http://ieeexplore.ieee.org/iel5/9010/28606/01294855.pdf?arnumber=1294855

http://en.wikipedia.org/wiki/Steganography

http://www.petitcolas.net/fabien/steganography/mp3stego/index.html