Binarization by Ohtsu's discriminant analysis method

binarize.py

from PIL import Image
import sys
import os

def binarize_try(hist, t):
  blist = hist[:t]
  wlist = hist[t:]
  if len(blist) == 0 or len(blist) == 0:
    return 0
  w1 = sum(blist)
  w2 = sum(wlist)
  m1 = w1 / len(blist)
  m2 = w2 / len(wlist)
  
  return (w1 * w2 * (m1 - m2) ** 2) / (w1 + w2) ** 2

def binarize(img):
  hist = img.histogram()
  t_conds = [binarize_try(hist, i) for i in xrange(0, 256)]
  
  t_max = 0
  t_max_idx = 0

  for i, t in enumerate(t_conds):
    if t > t_max:
      t_max_idx = i
      t_max = t
  print("t = {0}".format(t_max_idx))
  
  (img_w, img_h) = img.size
  pix = img.load()
  for x in xrange(0, img_w):
    for y in xrange(0, img_h):
      if pix[x, y] < t_max_idx:
        pix[x, y] = 0
      else:
        pix[x, y] = 255
  return img


if __name__ == '__main__':
  if len(sys.argv) != 2:
    print("usage: python binarize.py image_file")
    exit()
  img = Image.open(sys.argv[1])
  img = img.convert('L')
  img = binarize(img)
  img.save("{0}_bin.png".format(os.path.splitext(sys.argv[1])[0]), "png")