make_datasets

import numpy as np
import os, random
import cv2
import glob
import threading, multiprocessing
from PIL import Image, ImageDraw, ImageFont

image_size = 128
char_used_path = './chars.txt'
datasets_dir = './char_datasets'


def load_font(file, font_size=128):
    return ImageFont.truetype(open(file, 'rb'), size=font_size, encoding='utf-8')


def plot_char(canvas_size, offset, char, font):
    img = Image.new('L', (canvas_size, canvas_size), (255))
    draw = ImageDraw.Draw(img)
    draw.text((offset, offset), char, (0), font=font)
    return np.array(img)


def load_chars(chars_path):
    with open(chars_path, encoding='utf-8') as file:
        chars = file.read().strip()
        return list(chars)


def deal(files, start_index, chars):
    for i, file in enumerate(files):
        try:
            font = load_font(file, 100)
            images = [plot_char(128, 5, char, font) for char in chars]
            for j, img in enumerate(images):
                save_dir = datasets_dir + '/%04d/%04d.png' % (j, i + start_index)
                cv2.imwrite(save_dir, img)
        except Exception:
            print('font load failed:', file)
            continue


if __name__ == '__main__':
    if not os.path.exists(datasets_dir):
        os.mkdir(datasets_dir)
    normal_fonts = './fonts/normal_fonts'
    
    chars_patch = load_chars(char_used_path)
    chars_len = len(chars_patch)
    files = sorted(glob.glob(os.path.join(normal_fonts, '*.ttf')))
    files = [file.strip() for file in files]
    for i in range(chars_len):
        os.mkdir(datasets_dir + '/%04d' % i)
    for th in range(16):
        end_f = (th + 1) * 175
        if end_f > len(files):
            end_f = len(files)
        file_used = files[th * 175 : end_f]
        t = threading.Thread(target=deal, 
                             args=(file_used, th * 175, chars_patch))
        t.start()