ocr数据不够，怎么造数据

1.确定特定字体类型；

2.收集合适的图片作为背景

3.在背景图上填写特定字体的字符内容

1）字体无法确认时怎么办？

方法一：可以将文本行裁剪出来去网站上确认，网站链接：字体识别-在线扫一扫图片找字体-搜字体!

方法二:将文字输入到文档文件中，更换不同的字体，看是否与字体目标匹配；

字体可以去网上下载，也可以在本机查找；本机的字体所在位置：

个人用户字体文件：~/.local/share/fonts
系统字体文件：/usr/share/fonts
字体配置文件：/etc/fonts/

下面是我处理的代码，仅供参考：

def check_dir1(path):
    if not os.path.exists(path):
        os.mkdir(path)
    else:
        files = os.listdir(path)
        for file in files:
            file_path = os.path.join(path, file)
            os.remove(file_path)

'''
制作一些文本行数据
'''
from PIL import ImageFont, ImageDraw
import PIL.Image as PImage
import random
import os
import numpy as np
import cv2
from rec.temporary_boundary.line_process import cut_line3_1
from result_process.preprocess import check_dir1
 
if __name__=='__main__':
    cha_list = ['A','B','C','D','E','F','G','H','I','J','K',\
                'L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z']
 
 
    save_dir = '/home/fuxueping/4tdisk/data/certificate_reader/北京现场测试数据/20240614针对识别问题/SAU_name'
    check_dir1(save_dir)
    txt_parh = '/home/fuxueping/4tdisk/data/certificate_reader/北京现场测试数据/20240614针对识别问题/SAU_name.txt'
    bg_img_dir = '/home/fuxueping/4tdisk/data/certificate_reader/北京现场测试数据/20240614针对识别问题/bg'
    bg_imgs = os.listdir(bg_img_dir)
 
    f_save = open(txt_parh, 'w', encoding='utf-8')
    check_dir1(save_dir)
    num = 50
    while num:
        all_num = 0
        bg_img = random.choice(bg_imgs)
        num1=random.choice([2, 3])
        chr_str = ''
        all_num += num1
        while num1:
            chr_ = random.choice(cha_list)
            chr_str += chr_
            num1 -=1
 
        char_med = ''
        for i in range(3):
            num2=random.choice([5,6,7,8])
            chr_str2=''
            all_num += num2
            while num2:
                chr_ = random.choice(cha_list)
                chr_str2 += chr_
                num2 -= 1
            if i == 0:
                char_med += chr_str2+', '
            elif i == 1:
                char_med += chr_str2 + ' '
            elif i == 2:
                char_med += chr_str2 + ' '
 
        chr_1 = random.choice(cha_list)
 
        result_str = chr_str+' '+char_med+chr_1
        all_num += 1
 
 
 
 
        im = PImage.open(os.path.join(bg_img_dir, bg_img))
        w, h = im.size
 
        font_size = 24
        w_len = int(0 + all_num * (font_size-3) + 4)
        if w_len > w:
            num -= 1
            continue
        name_font = ImageFont.truetype('/home/fuxueping/4tdisk/data/certificate_reader/北京现场测试数据/20240614针对识别问题/fonts/n019003l.pfb', font_size)
        draw = ImageDraw.Draw(im)
        y_len = random.randint(0, h-font_size-5)
        color = tuple([random.randint(0, 20) for _ in range(3)])
        draw.text((2, y_len), result_str, fill=color, font=name_font)
 
        box = (0, y_len, w_len, y_len+font_size+5)
        rect_img = im.crop(box)
        image_array = np.array(rect_img)
        cv2_image = cv2.cvtColor(image_array, cv2.COLOR_RGB2BGR)
        result, _ = cut_line3_1(cv2_image)
        if len(result):
            region_rec = cv2_image[result[1]:result[3], result[0]:min(w, result[2]+2)]  # 裁剪出待识别的区域
            image_array = cv2.cvtColor(region_rec, cv2.COLOR_BGR2RGB)
            rect_img = PImage.fromarray(image_array)
        # image_array = cv2.cvtColor(cv2_image, cv2.COLOR_BGR2RGB)
        # rect_img = PImage.fromarray(image_array)
 
        save_path = os.path.join(save_dir, str(num)+'_'+result_str+'.jpg')
        line = save_path+'\t'+result_str+'\n'
        f_save.write(line)
        rect_img.save(save_path)
        num -= 1
    f_save.close()

# 根据设定的阈值和图片直方图，找出波峰，用于分隔字符
def find_waves_row(threshold, histogram):#行数是59
    # up_point = -1  # 上升点
    # is_peak = False
    # if histogram[0] >= threshold:
    up_point = 0 #起始位置
    is_peak = True
    wave_peaks = []
    top_cut = []
    for i, x in enumerate(histogram): #x是对应的像素和，i是行
        if is_peak and x >= threshold:
            if i - up_point >=2 :
                # top_cut.append((up_point, i)) #加这一行，相当于裁减掉多于的空行
                up_point = i-1
            else:
                up_point = i
            is_peak = False
        elif not is_peak and x < threshold:#随后找到字符消失的位置
            is_peak = True
            if 1 < i < histogram.shape[0]-1:#行数不是在开头也不在结尾
                wave_peaks.append((up_point, i+1))
            else:
                wave_peaks.append((up_point, i))
            up_point = i
 
    # if is_peak and up_point != -1 and i - up_point > 4:
    #     wave_peaks.append((up_point, i))
    if not is_peak and x >= threshold:#虽然数据已经结束，但是没有出现小于阈值的情况
        wave_peaks.append((up_point, i))
 
    return wave_peaks
 
def cut_line3_1(rgb_img, kernel_size = 3, y_len = 5, row_threshold=255 * 1, col_thresh = 255*1):
    '''
    切割出每一行，只保留高度满足条件的一行内容，然后切除掉每一行的前端后尾端的空白
    '''
    rgb_img = method_9(rgb_img) #高斯滤波
    # 使用sauvola进行二值化
    h, w = rgb_img.shape[:2]
    sau_bin = sauvola_bin(rgb_img) #sauvola二值化
    # cv2.imwrite('./../temp/sauvola_bin.jpg', sau_bin)
    # sau_bin = get_charcter_region(rgb_img)  # 局部区域算阈值二值化
    # cv2.imwrite('./../temp/sau_bin1.jpg', sau_bin)
    sau_bin_inv = 255 - sau_bin
    # cv2.imwrite('./../temp/sau_bin_inv1.jpg', sau_bin_inv)
    if kernel_size != 0:
        sau_bin_inv = cv2.medianBlur(sau_bin_inv, kernel_size)
    # cv2.imwrite('./../temp/sau_bin_inv_dinose1.jpg', sau_bin_inv)
 
    col_histogram = np.sum(sau_bin_inv, axis=1)
    wave_peaks = find_waves_row(col_thresh, col_histogram)
    result = []
 
    #找出高度最大的区域,只保留一行内容
    max_y = 0
    result_y = []
    if not len(wave_peaks):
        return [], sau_bin_inv
 
    for i, wave_peak in enumerate(wave_peaks):
        y1 = wave_peak[0]
        y2 = wave_peak[1]
 
        if y2 - y1 < y_len: #20之前是这个阈值 ，将高度不满足>=5的字符区域去掉
            continue
        if max_y < y2 - y1:
            max_y = y2 - y1
            result_y = [y1, y2]
 
    if len(result_y): #有时候裁剪的图片可能是没有字符，这种情况多出现在证件类别错误的情况
        y1 = result_y[0]
        y2 = result_y[1]
    else:
        return [], sau_bin_inv
 
    line_img = sau_bin_inv[y1:y2, :]
    # line_img_bgr = rgb_img[wave_peak[0]:wave_peak[1], :]
    # save_other = os.path.join(save_path, file + '_'+str(i)+'.jpg')
    # cv2.imwrite(save_other, line_img)
 
    row_histogram = np.sum(line_img, axis=0)  # 数组的每一列求和
    # row_max = np.max(row_histogram)
    # row_threshold = row_max - 255*1
 
    wave_peaks_line = find_waves_col(row_threshold, row_histogram)
    # cv2.imwrite('./../temp/line_img.jpg', line_img)
 
    x1 = 0
    x2 = w
    result_ = []
    for wave_ in wave_peaks_line:
        len_x = wave_[1] - wave_[0]
        if len_x > 5:
            result_.append(wave_)
 
    if len(result_):  # 有时候朝水平投影内容消失了，就用【0，w】代替
        x1 = result_[0][0]
        x2 = result_[-1][1]
 
    return [x1, y1, x2, y2], sau_bin_inv