This is an experiment in seeing how far you can get using platform-provided packages such as GraphicsMagick, CoreImage, etc. to provide a PIL-like interface but taking advantage of their support for more advanced features such as threading, broader format support (including JPEG-2000), vectorization, etc.
The goal is simple: a user should be able to install NativeImaging and do something like this to a program which is currently using PIL:
from NativeImaging import get_image_class
Image = get_image_class("GraphicsMagick")
Very fast JPEG 2000 thumbnail generation compared to GraphicsMagick. Requires the non-OSS AWARE library: http://www.aware.com/imaging/jpeg2000.htm
Currently supports typical web application usage: loading an image, resizing it and saving the result. Testing reveals mixed results, beating PIL when producing thumbnails from large TIFFs and underperforming when thumbnailing equivalent JPEGs, both by about 2:1.
Both CPython and PyPy are supported, with PyPy seeing performance gains using the CFFI backend instead of ctypes. Significant optimization gains are likely possible, particularly where the I/O functions marshall data in and out of the non-filename-based APIs where data is currently being copied.
Currently supports basic usage: loading an image, resizing it, and saving the result. Performance is generally quite decent as the Java Advanced Imaging API is quite tuned, if somewhat baroque in design.