ocropus-linetrain

#!/usr/bin/python
import code,pickle,sys,os,re
from optparse import OptionParser
from pylab import *
import ocropy
import traceback
from ocropy import dbtables,ocrobook,linerec

parser = OptionParser("""
usage: %prog [options] .../.../010001.png ...
""")
parser.add_option("-m","--model",help="model file for alignment",default=None)
parser.add_option("-M","--newmodel",help="model class for training",default="latin")
parser.add_option("-o","--output",help="output file",default="linerec.pymodel")
parser.add_option("-F","--filelist",help="list of input files",default=None)

(options,args) = parser.parse_args()

if options.filelist is not None:
    assert len(args)==0
    args = open(options.filelist).readlines()
    args = [s[:-1] for s in args]

if len(args)<1:
    parser.print_help()
    sys.exit(0)

ion()
show()

if os.path.exists(options.output):
    print options.output,"exists; please remove"
    sys.exit(1)

# instantiate the line recognizer
recognizer = linerec.LineRecognizer()

# load the line recognizer for alignment
recognizer.load(options.model)

# now start the per-character training
recognizer.startTraining(model=options.newmodel)

# add each of the individual lines
for file in args:
    print
    print "###",file
    print
    image = ocropy.bytearray()
    text_file = re.sub(r'\.png','.gt.txt',file)
    if not os.path.exists(text_file):
        print "no transcript"
        continue
    text = open(text_file,"r").read()
    ocropy.read_image_gray(image,file)
    try:
        recognizer.addTrainingLine1(image,text)
    except:
        traceback.print_exc()
        print "skipping"
        continue

print
print "### finishing the training"
print

# finish training; this usually takes a long time
recognizer.finishTraining()

# finally, save the model to disk
recognizer.save(options.output)