Extracted text from image

I wrote a program that uses OCR to extract text from PDF and image files. It's not easier to use OCR because I'm talking to a friend about the automation of assignments sent in PDF, and I'm typing another friend's report into Word, so I wrote a program using OCR. I wondered if I should try it.

motion

1. Select PDF or image file Use Tkinter.filedialog.askopenfilenames to select PDF and JPG files. Allows you to select multiple files at once.
2. Convert PDF to image OCR cannot be performed directly from PDF, so convert it to an image once. I'm using poppler and pdf2image.
3. Extract text from image Extracts text from the image converted from PDF and the image selected in 1. Use tesseract and PyOCR.
4. Output as .txt If the selected file is hoge.jpg, it will be output as hoge.txt.

code

main.py

import os
import pyocr
import tkinter
from tkinter import filedialog
from pdf2image import convert_from_path
from PIL import Image


class UseOCR:

    def __init__(self):
        pyocr.tesseract.TESSERACT_CMD = '/usr/local/bin/tesseract'
        self.poppler_executable_path = '/usr/local/bin/'
        self.initialdir = '~/'
        self.extract_lang = 'jpn+eng'
        self.extension = [('pdf files', '*.pdf'),
                          ('jpeg file', '*.jpeg'),
                          ('jpg file', '*.jpg'),
                          ('png file', '*.png')]

    def askfilenames(self):
        root = tkinter.Tk()
        root.withdraw()
        path = filedialog.askopenfilenames(filetypes=self.extension, initialdir=self.initialdir)
        return path

    @staticmethod
    def get_fileinfo(path):
        basename = tuple(map(os.path.basename, path))
        fileinfo = dict(zip(basename, path))
        return fileinfo

    def pdf_to_image(self, pdf):
        image = convert_from_path(pdf, poppler_path=self.poppler_executable_path)
        return image

    def image_to_text(self, image):
        tool = pyocr.get_available_tools()[0]
        txt = tool.image_to_string(
            image,
            lang='jpn',
            builder=pyocr.builders.TextBuilder()
        )
        return txt


if __name__ == '__main__':
    OCR = UseOCR()
    path = OCR.askfilenames()
    fileinfo = OCR.get_fileinfo(path)
    for basename, path in fileinfo.items():
        filename, extension = os.path.splitext(basename)
        if extension == '.pdf':
            image = OCR.pdf_to_image(path)[0]
            txt = OCR.image_to_text(image)
        else:
            image = Image.open(path)
            txt = OCR.image_to_text(image)
        with open('./output/{}.txt'.format(filename), mode='w') as f:
            f.write(txt)

Description

I will explain using a PDF with the following sentences. (This image is a PDF exported to jpg and cropped.)

constructor

#pyocr's TESSERACT_Rewrite CMD to tesseract path. Location → which tesseract
pyocr.tesseract.TESSERACT_CMD = '/usr/local/bin/tesseract'

# convert_from_path()The path of poppler to assign to the argument of. Location → which pdfinfo
self.poppler_executable_path = '/usr/local/bin/'

#Directory when tkinter started
self.initialdir = '~/'

#Character to OCR
self.extract_lang = 'jpn+eng'

#Specifying the extension to select with tkinter
self.extension = [('pdf files', '*.pdf'),
                  ('jpeg file', '*.jpeg'),
                  ('jpg file', '*.jpg'),
                  ('png file', '*.png')]

askfilenames Returns a tuple of the full path of the file selected by Tkinter.

>>> path = OCR.askfilenames()
>>> path
('/Users/Username/Desktop/hoge.pdf',)

get_fileinfo Taking a full-path tuple as an argument returns a full-path dictionary with the filename.

>>> fileinfo = OCR.get_fileinfo(path)
>>> fileinfo
{'hoge.pdf': '/Users/Username/Desktop/hoge.pdf'}

pdf_to_image If you pass the path of the PDF file as an argument, it returns a list of PIL Image objects. Since pdf2image and PyOCR have a dependency on Pillow, it is easier to handle if you return an Image object instead of making it an image file.

>>> for k,v in fileinfo.items():
...     image = OCR.pdf_to_image(v)
>>> image
[<PIL.PpmImagePlugin.PpmImageFile image mode=RGB size=1654x2339 at 0x10E1749E8>]

image_to_text It is the central process for performing OCR. If you pass an image file or Image object as an argument, OCR is performed and the text is returned.

>>> txt = OCR.image_to_text(image[0])
>>> txt
'Test test character 0123'