PPOCRLabel半自动工具标注自制身份证数据集

#导入所需要的包from PIL import Image, ImageDraw, ImageFontimport numpy as npimport cv2import random

class Person(object):
    def __init__(self, name, sex, national,
                years, month, day, address,
                id_card):
        self.name=str(name), 
        self.sex=str(sex), 
        self.national=str(national),
        self.years=str(years), 
        self.month=str(month), 
        self.day=str(day), 
        self.address=str(address),
        self.id_card=str(id_card)

def GBK2312():
    """
    功能 : 随机生成一个汉字
    """
    head = random.randint(0xb0, 0xf7)
    body = random.randint(0xa1, 0xf9)  # 在head区号为55的那一块最后5个汉字是乱码,为了方便缩减下范围
    val = f'{head:x}{body:x}'
    st = bytes.fromhex(val).decode('gb2312')    return stdef first_name():    
    """
    功能 : 随机取姓氏字典
    """
    first_name_list = [        '赵', '钱', '孙', '李', '周', '吴', '郑', '王', '冯', '陈', '褚', '卫', '蒋', '沈', '韩', '杨', '朱', '秦', '尤', '许',        '何', '吕', '施', '张', '孔', '曹', '严', '华', '金', '魏', '陶', '姜', '戚', '谢', '邹', '喻', '柏', '水', '窦', '章',        '云', '苏', '潘', '葛', '奚', '范', '彭', '郎', '鲁', '韦', '昌', '马', '苗', '凤', '花', '方', '俞', '任', '袁', '柳',        '酆', '鲍', '史', '唐', '费', '廉', '岑', '薛', '雷', '贺', '倪', '汤', '滕', '殷', '罗', '毕', '郝', '邬', '安', '常',        '乐', '于', '时', '傅', '皮', '卞', '齐', '康', '伍', '余', '元', '卜', '顾', '孟', '平', '黄', '和', '穆', '萧', '尹',        '姚', '邵', '堪', '汪', '祁', '毛', '禹', '狄', '米', '贝', '明', '臧', '计', '伏', '成', '戴', '谈', '宋', '茅', '庞',        '熊', '纪', '舒', '屈', '项', '祝', '董', '梁', '魏']
    n = random.randint(0, len(first_name_list) - 1)
    f_name = first_name_list[n]    return f_namedef second_name():
    """
    功能 : 随机取数组中字符，取到空字符则没有second_name
    """
    second_name_list = [GBK2312(), '']
    n = random.randint(0, 1)
    s_name = second_name_list[n]    return s_namedef last_name():
    """
    功能 : 随机生成名字最后一位字
    """
    return GBK2312()def create_name():
    """
    功能 : 随机生成名字
    """
    name = first_name() + second_name() + last_name()    return namedef sex_word():
    """
    功能 : 随机生成性别
    """
    nums = random.randint(0,3)
    sex_list = [        '男', '女', '机器人', '未知']    
    return sex_list[nums]def address_name():  
    """
    功能 : 随机生成省份
    """
    first_name_list = [        '黑龙江省哈尔滨', '吉林省长春', '辽宁省沈阳', '河北省石家庄', '山西省太原', '青海省西宁', '山东省济南', '河南省郑州', '江苏省南京', '安徽省合肥',        '浙江省杭州', '福建省福州', '江西省南昌', '湖南省长沙', '湖北省武汉', '广东省广州', '台湾省台北', '海南省海口', '甘肃省兰州']
    n = random.randint(0, len(first_name_list) - 1)
    address_name = first_name_list[n]    return address_namedef random_id_card():
    """
    功能 : 随机生成18位身份证ID
    """
    num_str = ''
    _rand = random.randint(0, 100)    for i in range(17): #  
        # num_str=str(random.randint(0, 9)).zfill(17)
        num_str = num_str + str(random.randint(0, 9))    if _rand >= 92:
        num_str = num_str + 'X'
    else:
        num_str = num_str + str(random.randint(0, 9))    
    return num_strdef to_str(per):
    """
    功能 : 将元祖转为str
    """
    _str =''.join(per)    
    return _str

import random

def Unicode():    name = ""
    nums = random.randint(2,4)    for n in range(nums):
        val = random.randint(0x4e00, 0x9fbf)        name = name + chr(val)    return namename = Unicode()name

# 定义写字函数def add_txt(image, size, draw_x, draw_y, txt):
    # 字体字号
    setFont = ImageFont.truetype('IDTemplate/simhei.ttf', size)    # 定义画板
    draw = ImageDraw.Draw(image)    # 绘制
    draw.text((draw_x, draw_y), txt, font=setFont, fill=(0, 0, 0))    return image 
 
def make_fake_id_card(person):
    
    
    ori_image = cv2.imread('./IDTemplate/IDCard{}.png'.format(random.randint(0,8))) 
    print('==ori_image.shape:', ori_image.shape)
    ori_image = cv2.resize(ori_image, (0, 0), fx=0.4, fy=0.4)    print('==resize ori_image.shape:', ori_image.shape)    
    
    # 向图片上写字
    img = Image.fromarray(cv2.cvtColor(ori_image, cv2.COLOR_BGR2RGB))
    img = add_txt(img, 19, 97, 58, to_str(person.name)) #字体 x坐标 y坐标
    img = add_txt(img, 16, 97, 90, to_str(person.sex))
    img = add_txt(img, 16, 190, 90, to_str(person.national))
    img = add_txt(img, 16, 92, 118, to_str(person.years))
    img = add_txt(img, 16, 162, 118, to_str(person.month))
    img = add_txt(img, 16, 201, 118, to_str(person.day))
    img = add_txt(img, 16, 94, 155, to_str(person.address))
    img = add_txt(img, 16, 94, 176, '某某　666号')
    img = add_txt(img, 18, 147, 239, to_str(person.id_card))
 
    cv2.imwrite('./train_data/middleIMG/word.jpg', np.array(img)[..., ::-1]) 
def make_white_mask(person, nums):
    # 生成一个空白的模板mask
    ori_image = cv2.imread('./IDTemplate/IDCard{}.png'.format(random.randint(0,8)))
    ori_image = cv2.resize(ori_image, (0, 0), fx=0.4, fy=0.4)
    mask_image = np.ones_like(ori_image)
    mask_image *= 255
    print(mask_image.shape,' {}.jpg'.format(nums))
    cv2.imwrite('./train_data/middleIMG/mask.jpg', mask_image) 
    # 往空白模板上写字(这里只能用PIL写，因为OpenCV写中文会乱码)
    img = Image.fromarray(cv2.cvtColor(ori_image, cv2.COLOR_BGR2RGB))
    img = add_txt(img, 19, 97, 58, to_str(person.name)) #字体 x坐标 y坐标
    img = add_txt(img, 16, 97, 90, to_str(person.sex))
    img = add_txt(img, 16, 190, 90, to_str(person.national))
    img = add_txt(img, 16, 92, 118, to_str(person.years))
    img = add_txt(img, 16, 162, 118, to_str(person.month))
    img = add_txt(img, 16, 201, 118, to_str(person.day))
    img = add_txt(img, 16, 94, 155, to_str(person.address))
    img = add_txt(img, 16, 94, 176, '某某　666号--自制数据集')
    img = add_txt(img, 18, 147, 239, to_str(person.id_card))
 
    mask_image_txt = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
    cv2.imwrite('./train_data/middleIMG/mask_image_txt.jpg', mask_image_txt)
    gray = cv2.cvtColor(mask_image_txt, cv2.COLOR_BGR2GRAY)    # 高斯模糊，制造边缘模糊效果哦
    gray_Gaussianblur = cv2.GaussianBlur(gray, (3, 3), 0) 
    # 使用阈值对图片进行二值化
    th, res = cv2.threshold(gray_Gaussianblur, 200, 255, cv2.THRESH_BINARY)
    res_inv = cv2.bitwise_not(res)
    cv2.imwrite('./train_data/middleIMG/res_inv.jpg', res_inv) 
    # 写字的模板保留文字部分
    img_bg = cv2.bitwise_and(mask_image_txt, mask_image_txt, mask=res_inv)
    cv2.imwrite('./train_data/middleIMG/img_bg.jpg', img_bg)    # 原图保留除文字的其他部分
    img_fg = cv2.bitwise_and(ori_image, ori_image, mask=res)
    cv2.imwrite('./train_data/middleIMG/img_fg.jpg', img_fg)    # 将两张图直接进行相加，即可
    final = cv2.add(img_bg, img_fg)
    cv2.imwrite('./train_data/{}.jpg'.format(nums), final)

!mkdir -p train_data/middleIMG

if __name__ == '__main__':    # make_fake_id_card()
    for i in range(3200):
        person = Person(name=create_name(), sex=sex_word(), national='汉',
                    years=str(random.randint(1970, 2020)), month=str(random.randint(1, 12)), 
                    day=str(random.randint(1, 28)), address=address_name(), id_card=random_id_card())
        make_white_mask(person, str(i).zfill(4))

#解压zip !zip -q -r train_data.zip train_data/

   1.中文与英文模型权重 
    
   2.英文模型权重 
   
   3.法语模型权重 
   
   4.德语模型权重 
   
   5.韩文模型权重 
   
   6.日文模型权重