Tiffany - Read/Write Multipage-Tiff with PIL without PIL

Tiffany stands for any tiff. The tiny module solves a large set of problems, has no dependencies and just works wherever Python works. Tiffany was developed in the course of the Pydica project and will now appear on PyPi.


During the development of Pydica (Python Distributed Capture) we were confronted with the problem to read multipage Tiff scans. The GUI toolkit PySide (Qt) does support Tiff, but only shows the first page. We also had to support Fax compression (CCITT G3/G4), but Qt supports this.

As a first approach we copied single pages out of multi-page tiff files using tiffcp or tiffutil (OS X) as a temp file for display. A sub-optimum solution, especially for data security reasons.

The second approach replaced this by a tiny modification of the linkage of the tiff directories (IFD). This way, a tiff file could be patched in memory with the wanted page offset and then be shown without any files involved.

Unfortunately also this solution was not satisfactory:

  • our tiff files have anomalies in their tiff tags like too many null-bytes and wrong tag order,
  • Qt's implementation of tiff is over-pedantic and ignores all tags after the smallest error.

Being a good friend of Fredrik Lundh and his PIL since years, I tried to attack the problem using this. Sadly Fredrik hasn't worked much on this since 2006, and the situation is slightly messed up:

PIL has a clean-up of tiff tags, but cannot cope with fax compression by default. There exists a patch since many years, but this complicates the build process and pulls with libtiff a lot of dependencies in.

Furthermore, PIL is unable to write fax compressed files, but blows the data up to the full size, making this approach only a half solution as well.

After a longer odyssey I saw then the light of a Tiffany lamp:

I use only a hand-full of *PIL*s files, without any modification, pretend to unpack a tiff file, but actually cheating. Only the tiff tags are nicely processed and streamlined, but the compressed data is taken unmodified as-is. When writing a tiff page out, the existing data is just assembled in the correct order.

For many projects like Pydica that are processing tiff files without editing their contents, this is a complete solution of their tiff problem. The dependencies of the project stay minimal, there are no binaries required, and Tiffany is with less than 300 lines remarkably small.

Because just 5 files from PIL are used and the _imaging module is not compiled at all, I'm talking about "PIL without PIL" ;-)

Tiffany is a stand-alone module and has no interference with PIL. You can see this by looking at import_mapper.py. This module modifies __import__ so that the PIL modules appear as top-level internally, but become sub-modules of tiffany in sys.modules.

Please let me know if this stuff works for you, and send requests to <tismer@stackless.com> or use the links in the bitbucket website:


easiest way to install tiffany:

$ pip install tiffany