从这节开始，就正式进入GCN项目的代码环节，我们先使用PaddleOCR，识别出火车票上的文字信息，并提取文字所位置，整理后另存为一个csv文件。

项目的目的是要做信息抽取，所以对训练和测试数据，需要手动打上标签，供算法学习。

代码示例

1、新建文件

# process/ocr.py
from paddleocr import paddleocr, PaddleOCR
import pandas as pd
from glob import glob
import os
import cv2
from tqdm import tqdm
import logging

# 屏蔽调试错误
paddleocr.logging.disable(logging.DEBUG)

2、定义OCR识别类

主要功能是识别信息并存储csv文件，如果需要导出识别框标记的图片，则可以传入marked_path。

class OCR():
    def __init__(self):
        self.ocr = PaddleOCR()

    def scan(self, file_path, output_path, marked_path=None):
        # 文字识别
        info = self.ocr.ocr(file_path, cls=False)
        df = pd.DataFrame(columns=['x1', 'y1', 'x2', 'y2', 'text'])
        for i, item in enumerate(info[0]):
            # 保留左上和右下坐标
            ((x1, y1), _, (x2, y2), _), (text, _) = item
            df.loc[i] = list(map(int, [x1, y1, x2, y2])) + [text]
        # 保存识别结果
        df.to_csv(output_path)
        # 判断是否需要保存标记文件
        if marked_path:
            self.marked(df, file_path, marked_path)

    # 导出带标记的图片
    def marked(self, df, file_path, marked_path):
        # 加载图片
        img = cv2.imread(file_path)
        for x1, y1, x2, y2, _ in df.values:
            # 画矩形（注意坐标值必须为整数）
            cv2.rectangle(img, (x1, y1), (x2, y2), color=(0, 0, 255), thickness=4)
        cv2.imwrite(marked_path, img)

3、识别测试和训练图片

if __name__ == '__main__':
    ocr = OCR()
    for file_path in tqdm(glob('../input/imgs/train/' + '*.*')):
        _, file_name = os.path.split(file_path)
        output_path = '../output/train/csv/' + file_name + '.csv'
        marked_path = '../output/train/imgs_marked/' + file_name
        ocr.scan(file_path, output_path, marked_path)

    for file_path in tqdm(glob('../input/imgs/test/' + '*.*')):
        _, file_name = os.path.split(file_path)
        output_path = '../output/test/csv/' + file_name + '.csv'
        marked_path = '../output/test/imgs_marked/' + file_name
        ocr.scan(file_path, output_path, marked_path)

4、手动打标签

a) 标签示例

,label
,ticket_num
,starting_station
,destination_station
,train_num
,other
,date
,seat_number
,ticket_price
,ticket_grade
,name

b) 标记后的csv文件示例

,x1,y1,x2,y2,text,label
0,814,41,1086,107,=检票：2,other
1,109,54,332,98,F049081,ticket_num
2,134,119,423,185,德州东站,starting_station
3,758,114,1053,184,温岭站,destination_station
4,530,128,660,183,G55,train_num
5,158,190,395,233,Dezhoudong,other
6,826,183,982,235,Wenling,other
7,94,247,634,289,2017年10月21日09:26开,date
8,741,244,962,288,06车08C号,seat_number
9,485,302,583,353,网折,other
10,822,300,960,350,二等座,ticket_grade
11,97,310,308,351,￥541.5元,ticket_price
12,98,369,412,408,限乘当日当次车,other
13,101,486,736,528,1329271981****0518崔厚良,name
14,223,559,751,590,买票请到12306发货请到95306,other
15,289,607,681,641,中国铁路祝您旅途愉快,other
16,105,670,749,712,66809300121022F049081德州东售,other