Image classification with pybrain by Brian Cole
This was originally a project for class (CS 344, AI); it uses a neural network to classify images, given images to pattern off of. For instance, if you have directories full of images of goats and whales, and a directory full of images of assorted mammals:
.
./goats
./goats/babyGoat.jpg
./goats/rammingRam.png
./goats/fakeSheep.jpeg
./whales
./whales/bigBlue.png
./whales/greatWhite.jpg
./whales/helloGround.png
./unknown
./unknown/bigBlueThing.png
./unknown/whiteFurryThing.png
./unknown/brownFurryThing.png
./unknown/cat.png
./probablyGoats
./probablyWhales
Then you should be able to run the command
recognizer_cli.py goats whales unknown probablyGoats probablyWhales
and end up with the contents of unknown/ moved into probablyGoats and probablyWhales as is appropriate. The cat will probably end up with the goats, since the neural network will always say that an image is closer to one category or the other.
The main program here is recognizer_cli.py, but I've also written a stripped-down version that doesn't train a neural network, but instead loads one from a file (recognizernet.brain by default); this is an automatically-generated dump of the last neural network generated and trained by recognizer.py. In this way, one may train a network, then use it later to classify images without having to rebuild the network again. This internally uses python's pickle functionality.
To use this read-only version on the above example, one would run
recognizer_cli_read.py unknown probablyGoats probablyWhales
It's nearly the same thing, but without the training sets.