基于深度学习的验证码识别.zip

''' @Author: by @Date: 2019-07-08 14:04:11 @LastEditTime: 2019-07-12 11:06:08 ''' import argparse import os import string import cv2 import h5py def get_args(): parser = argparse.ArgumentParser(description="Convert captcha images") parser.add_argument("--path", metavar="PATH", default=None, help="Captcha images output path.") parser.add_argument("--out", metavar="OUT", default=None, help="Captcha images output path.") args = parser.parse_args() return args def convert(shape=(80, 170, 3), class_num=36, char_num=4): ''' Convert image and label to .h5py. :param input_dir: Images path. :param output_file: Output filename. :param shape: Input imaghe shape = (height, width, channel). :param class_num: Classex numbers. :param char_num: The number of every captch contains characters. :return: None ''' args = get_args() input_dir = args.path output_file = args.out if os.path.exists(output_file): os.remove(output_file) input_h, input_w, input_c = shape characters = string.digits + string.ascii_uppercase with h5py.File(output_file, 'w') as f: f.create_dataset("input", (1024, input_h, input_w, input_c), maxshape=(None, input_h, input_w, input_c), dtype='float32') f.create_dataset("label", (char_num, 1024, class_num), maxshape=(char_num, None, class_num), dtype='int32') img_list = os.listdir(input_dir) img_num = len(img_list) count = 0 h5f = h5py.File(output_file, 'a') for j, f in enumerate(img_list): print("\r%d/%d : %s" % (j, img_num, f), end="") if count >= h5f['input'].shape[0]: input_nums = h5f['input'].shape[0] * 2 h5f['input'].resize((input_nums, input_h, input_w, input_c)) h5f['label'].resize((char_num, input_nums, class_num)) full_path = os.path.join(input_dir, f) img = cv2.imread(full_path) img = img / 256. label = f.split(".")[0] h5f["input"][count] = img for i, _ in enumerate(label): h5f["label"][i, count, characters.find(_)] = 1 count += 1 h5f['input'].resize((count, input_h, input_w, input_c)) h5f['label'].resize((4, count, 36)) h5f.close() print("\nDone!") if __name__ == "__main__": convert()