Improve image processing pipeline backend

Study existing solutions. Relate to threads/multiprocessing/coroutines

http://pypi.python.org/pypi?:action=search&term=pipeline&submit=search

For example
Kamaelia
Kamaelia is a toolset and mindset aimed at assisting in structuring your code such that you can focus on the problem you want to solve, but in a way that results in naturally reusable code that happens to be painlessly concurrent. It was designed originally to make maintenance of highly concurrent network systems simpler, but has general application in a wider variety of problem domains, including desktop applications, web backend systems (eg video transcode & SMS services), through to tools for teaching a child to read and write.
http://www.kamaelia.org/PragmaticConcurrency

filterpype
FilterPype is being used for multi-level data analysis, but could be applied to many other areas where it is difficult to split up a system into small independent parts. Some of its features:
Advanced algorithms broken down into simple data filter coroutines
Pipelines constructed from filters in the new FilterPype mini-language
Domain experts assemble pipelines with no Python knowledge required
Sub-pipelines and filters linked by automatic pipeline construction
All standard operations available: branching, joining and looping
Recursive coroutine pipes allowing calculation of e.g. factorials
Using it is like writing a synchronous multi-threaded program
- http://www.europython.eu/talks/talk_abstracts/index.html#talk94
- http://www.foldingsoftware.com/filterpype/

others
- http://code.google.com/p/python-pipeline/

coroutines
- http://www.dabeaz.com/coroutines/

pipeline general
- gstreamer
- GEGL (image processing framework)