2009-08-05 19:38:50 UTC
Hello jesse .
The tool you are looking for is in
org.neuroph.contrib.tileclassification.gui.TileClassificationTrainerGui
This is the tool described at dota-script site. The support for image recognition in easyNeurons is under development, and it is based on this tool.
1. To start image recognition tool type:
java -cp easyneurons.jar org.neuroph.contrib.tileclassification.gui.TileClassificationTrainerGui
2.In the field 'Neural network name' Type the name for your neural network like MyNeuralNetwork.nnet
This is the name of file where your network will be saved
3. Put images you want to recognize in one directory, and give appropriate name for each image. Image names will be used as labels for outputs. Click 'Choose directory of labeled images' button and select that directory. Use 'Choose directory of junk images' to specify images that should not be recognized (for example full red, green and blue squares).
4. Click 'start learning button'. When Net error around 0.01 click Stop Learning.
5. To test network click 'Choose test Image' and select some image you used for training. In the text box you'll get the posibility that chosen image belongs to some of the categories.
This tool can be used to recognize 8x8 pixels color images. Bigger images must be scaled to this dimension.
Cheers,
Zoran