根据字库生成一个图片，图片里面约10多个字，白底黑字，字号大约6号：

 大概长这个样子

首先找个字库再说：https://github.com/elephantnose/characters

导入依赖包

import cv2
from PIL import ImageFont,ImageDraw,Image
import numpy as np
import uuid
import os
import random
import math

配置函数，画图函数，生成函数

def config():
    with open("word_dict2.txt") as f:
        word = f.readlines()
    word_dict_length = len(word)#字库字数
    fontPath = "PingFang Heavy.ttf"#黑体
    save_dir = "gen/images"
    min_words_length = 9
    max_words_length = 13
    img_count = 10
    return word,fontPath,save_dir,min_words_length,max_words_length,img_count


def draw_to_image(size,fontPath,font_size,text,save_dir,file_name):
    background = np.ones(size)*255
    font = ImageFont.truetype(fontPath,font_size)
    img = Image.fromarray(background)
    draw = ImageDraw.Draw(img)
    draw.text((1,1),text,font=font,fill='black')
    img = np.array(img)
    out_path = os.path.join(save_dir,file_name)
    cv2.imwrite(out_path,img)
    
    
def generate_words(word,count,length_range_min,length_range_max):
    line_words_list = []
    line_words = ""
    globe_word_dict_index = 0
    for k in range(count+1):
        bit_length = np.random.randint(length_range_min,length_range_max)
        for i in range(bit_length+1):
            try:
                line_words = line_words + word[globe_word_dict_index].split("\n")[0]
            except:
                globe_word_dict_index = 0
                line_words = line_words + word[globe_word_dict_index].split("\n")[0]
            globe_word_dict_index+=1
        line_words_list.append(line_words)
        line_words = ""
    return line_words_list

调用

word,fontPath,save_dir,min_words_length,max_words_length,img_count = config()
line_words_list = generate_words(word,img_count,min_words_length,max_words_length)
for line in line_words_list:
    length = len(line)
    font_size = 33 if np.random.randint(0,2)==0 else 20
    file_name = str(uuid.uuid4())+".jpg"
    if(font_size==33):
        size = (42,math.ceil(length*34.4))
        draw_to_image(size,fontPath,font_size,line,save_dir,file_name)
    else:
        size = (29,math.ceil(length*20.8))
        draw_to_image(size,fontPath,font_size,line,save_dir,file_name)
    with open("label","a") as f:
        f.write("images/"+file_name+"\t"+line+"\n")

最后这段是写出图像文件的同时把文件位置，文件名以及文字标签追加到本地文档

格式是PaddleOCR自定义数据集格式：

" 图像文件名                 图像标注信息 "

train_data/rec/train/word_001.jpg   简单可依赖
train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
...