The Python Imaging Library does not support handling of UTF-8 'iTXt' key:value chunks in PNG files:

  http://www.w3.org/TR/PNG/#11iTXt

Such support is necessary for successful extraction of key:value pairs of UTF-8 encoded data, stored in an PNG 'iTXt' comment chunk.

The following example file (from British GCHQ) demonstrates such a record in a PNG file.  Based on this evidence, it is highly likely that GCHQ hide all of their important secrets using this kind of steganography, and likely necessary that spies from the rest of the world are requiring similar access to GCHQ secrets.  Inclusion of a working chunk_iTXt() PIL/PNG support function will enable more harmonious and effective eavesdropping.  Example image:

  http://www.canyoucrackit.co.uk/images/cyber.png

(The attached .py file is not a directly apply-able patch, but contains a working implementation for chunk_iTXt() and a demonstrative test function for inserting it).