data-to-images.py

from PIL import Image
import numpy
import os

DIR = "images_good"

W, H = 16, 16
MNIST_NORM = (20, 20)
MNIST_SIZE = (28, 28)

WHITE = (255, 255, 255)

def tokens_to_img(tokens):
    np_arr = numpy.empty((W,H), numpy.uint8)
    #np_arr.shape=H,W
    
    for i in range(len(tokens) - 11):
        row = i // W
        col = i % H
        np_arr[row, col] = (1-float(tokens[i])) * 255
    
    return Image.fromarray(np_arr)

def tokens_to_num(tokens):
    le = len(tokens)
    num = 0
    for i in range(le - 11, le):
        if tokens[i] == '1':
            return num
        else:
            num = num + 1

def normalize_img(img):
    img = img.resize(MNIST_NORM)
    background = Image.new('RGB', MNIST_SIZE, WHITE)
    bg_w, bg_h = background.size
    img_w, img_h = img.size
    offset = ((bg_w - img_h) / 2, (bg_h - img_h) / 2)
    background.paste(img, offset)
    return background

if not os.path.exists(DIR):
    os.makedirs(DIR)
    
with open("semeion.data") as f:
    idx = 0
    for line in f:
        idx = idx + 1
        tokens = line.split(' ')
        img = tokens_to_img(tokens)
        img_normalized = normalize_img(img)
        num = tokens_to_num(tokens)
        filename = "{DIR}/image{idx}_{num}.png".format(DIR=DIR, idx=idx, num=num)
        img_normalized.save(filename)