You are here: Home / Documentation

Documentation

SOCR

The following instructions are for Epicurus-0.83, Epicurus-apps-1.03, and SOCR-0.31.

In the SOCR-0.31 directory build:

    make process_pgm

This will create the process_pgm command which allows you to build a model based on a series of pbm/pgm files and then make predictions.

At this early stage you can use it like:

    ./process_pgm -s test.model -t images/*

This will calculate the features for each of the images and create a Naive Bayes model for them all. If you have a test collection that you’d like predictions for, simply:

    ./process_pgm -l test.model -T images/*

NB: Watch the -t (train) -T (test/predict) case differences

Learning new fonts

If you want to use this simple command for learning a new font you’ll have to do two things:

  • Edit SOCR_features.cc and add the name of your character to the list of supported characters, ie. ka, or sa
  • Create a series of ka.1, ka.2, ka.3 files (the number is not used, but you need it for unique filenames). These files must be in .pgm or .pbm format. Use the # character in the files for storing extra information that may be useful later. eg. 
    	  # resolution: 300
	  # source: IEEE
	  # baseline: 12