Home > Article > Backend Development > Tips | Python automatically extracts and organizes PDF invoices in batches
This article shares a PDF-based Python office automation case solution, which was also proposed by a financial lady. Let’s first look at the real needs.
There are multiple PDF type invoices in a certain folder
Each one Invoice PDF is a pure picture type, and the text information inside cannot be copied manually (in fact, most invoices can copy part of the text, but we will explain it in the form of pictures), as shown below:
The requirements that need to be met are: obtain the total amount of , taxpayer identification number, and issuer , that is, the following three box positions:
Finally combined with the batch Operation, after obtaining the above information, store it in Excel!
The essence of the requirement is an image recognition problem, because the content in the PDF is of image type and cannot be pressed Conventional methods extract the text directly. The solution is to use optical character recognition (OCR) to recognize the text in the picture. But at the same time, it should be noted that PDF is not a picture after all. In order to complete OCR, in addition to OCR itself, Ghostscript
and ImageMagick
must be downloaded to complete type conversion. Taking the Windows
system as an example, you need to install the following three software on your computer:
Ghostscript
32-bitImageMagick
32-bittesseract-OCR
32-bitThere is no special place in downloading and installing the three software (tesseract configuration is slightly complicated, but there are many tutorials on the Internet, so I won’t go into details here. ), readers can search for download and configuration by themselves. The code is explained below. First import the required modules:
from wand.image import Image from PIL import Image as PI import pyocr import pyocr.builders import io import re import os import shutil
For specific module uses, please refer to the specific code below. Among them, wand
and pyocr
need to be installed separately because they are non-standard libraries. Open the command line and enter:
pip install wand pip install pyocr
This requirement also involves docking Excel. You can consider using the Workbook
of the openpyxl
library to create a new Excel file:
from openpyxl import Workbook
Put the invoice.pdf
in the requirement on the desktop. The desktop path can be obtained through the following code based on the os
module:
# 获取桌面路径包装成一个函数 def GetDesktopPath(): return os.path.join(os.path.expanduser("~"), 'Desktop') path = GetDesktopPath() + r'\发票.pdf'
Get the configured tesseract
for later calling:
tool = pyocr.get_available_tools()[0]
通过 wand
模块将 PDF 文件转化为分辨率为 300 的 jpeg
图片形式:
image_pdf = Image(filename=path, resolution=300) image_jpeg = image_pdf.convert('jpeg')
将图片解析为二进制矩阵:
image_lst = [] for img in image_jpeg.sequence: img_page = Image(image=img) image_lst.append(img_page.make_blob('jpeg'))
用 io
模块的 BytesIO
方法读取二进制内容为图片形式:
new_img = PI.open(io.BytesIO(image_lst[0])) new_img.show()
接下来分别截取需要提取部位字符串的图片了,尽量让图片中只有需要识别的部分,获取识别出来容易简单处理获得需要的内容。
首先以总金额为例,截取图片用 image.crop((left, top, right, bottom))
四个参数需要反复调试才能确定。经确定四个参数分别是 1600 760 1830 900,尝试截取和预览图片:
### 解析1Z开头码 left = 350 top = 600 right = 1300 bottom = 730 image_obj1 = new_img.crop((left, top, right, bottom)) image_obj1.show()
截取成功后可以交给 OCR 了,代码为 tool.image_to_string()
txt1= tool.image_to_string(image_obj1) print(txt1)
同样,通过方位的调试就可以准确切割到需要的部分进行识别:
left = 560 top = 1260 right = 900 bottom = 1320 image_obj2 = new_img.crop((left, top, right, bottom)) # image_obj2.show() txt2 = tool.image_to_string(image_obj2) # print(txt2)
最后是开票人的识别
left = 1420 top = 1420 right = 1700 bottom = 1500 image_obj3 = new_img.crop((left, top, right, bottom)) # image_obj3.show() txt3 = tool.image_to_string(image_obj3) # print(txt3)
需要确认识别的内容是否正确,如果识别正确率欠佳可以考虑通过图片处理技术消除噪声,也可以去官网下载更高精度的训练包提高识别的正确性
至此,我们成功的识别了总金额、纳税人识别号、开票人三个消息,接下来就通过非常熟悉的 openpyxl
写入Excel,并使用 os
模块实现批量操作即可
workbook = Workbook() sheet = workbook.active header = ['总金额', '纳税人识别号', '开票人'] sheet.append(header) sheet.append([txt1, txt2, txt3]) workbook.save(GetDesktopPath() + r'\汇总.xlsx')
综上,整个需求就成功实现,从效果来看还是非常不错的!完整源码可由文中代码组合而成(已全部分享在文中),感兴趣的读者可以自己尝试!
最后想说的是,其实本文的案例可以衍生出很多实用的办公自动化脚本,例如
The above is the detailed content of Tips | Python automatically extracts and organizes PDF invoices in batches. For more information, please follow other related articles on the PHP Chinese website!