handle_scanned_pdf

A wrapper on top of python-OCR tools such as pytesseract and easyocr, to recognize and extract text embedded in images. The first open-source wrapper package that converts scanned PDF files to searchable PDF files (end-to-end) facilitates using EasyrOCR and tesseract OCR.

Source code can be accessed here sxaxmz/handdle_scanned_pdf

Install the package using:

$ pip install handle-scanned-pdf

Server Installation

$ apt-get install poppler-utils

---

Features:

• Convert scanned-PDFs to text searchable PDFs (end-to-end).
• Extract text from scanned PDFs and images.
• Draw bounding boxes around the text that can be extracted on scanned PDFs and images.
• Recognize and extract text in various languages.
• The searchable PDF output places the extracted text and positions it accordingly on top of the inputted file.
• Ability to use one OCR to create a searchable PDF and a different OCR to extract text files (separately).
• If it is only desired to use EasyOCR, then Tesseract installation is not required.

Usage:

• Make scanned documents searchable and parsable.
• Helpful in digitizing archives.
• Make use of scanned documents and images when it's intended to be used for RAG applications.

Challenges:

• OCR performance and accuracy may vary based on the type of inputted data.
• The text position on the custom searchable PDF created (using easyocr) might not be 100% accurate as it takes the mean of top-right and bottom-right to produce x and y.

Feel Free to contribute to this repository.

---

Tesseract-OCR supports:

• Various image types including (but not limited to) jpeg, png, gif, bmp, tiff.
• Wide range of languages list of languages
• Supports reading more than 1 language at a time.

Server Installation

$ apt install tesseract-ocr

Only if required set the below path to Tesseract executable:

pytesseract.pytesseract.tesseract_cmd = r'/usr/bin/tesseract/tesseract.exe'

EasyOCR supports:

• Does not require server installation.
• Supports reading more than 1 language at a time.
• Performes faster on a GPU.
• List of supported language code.

Language Support

Tesseract

• Ensure to download the right Tesseract-OCR for the language needed to be used.

Installation on Linux:

$ apt install tesseract-ocr-<language-code>

Download for Windows (set path to the downloaded OCR):

• Download language files
• Add the folder that contains the downloaded files into the System Path Variables as TESSDATA_PREFIX

Defining the language codes for Tesseract:

lang_code = "eng+ara"
txt_extract_lang_code = "eng+ara"

EasyOCR

Defining the language codes for EasyOCR:

lang_code = ["en","ar"]
txt_extract_lang_code = ["en","ar"]

Packages Required (src: requirements.txt):

pytesseract===0.3.10
pdf2image===1.17.0
PyPDF2===3.0.1
opencv-python

Easy-to-Use:

• Straightforward functions.
• Customizable process.
• JSON output.

Draw bounding boxes on the text that can be extracted from PDF

from handle_scanned_pdf import draw_bounding_boxes

img_path = 'sample__images/3ba4c1f1-775f-4e05-ab48-a40617087a57-1.png'
img = np.array(cv2.imread(img_path)) # Read image and convert to numpy array
output_path = 'output'
file_name = os.path.basename(img_path).split('.')[0]
pageNum = 0
draw_bounding_boxes(img, output_path, file_name, pageNum)

Output:

output/images_bounding/3ba4c1f1-775f-4e05-ab48-a40617087a57-1_bounding_images/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_0.jpg

Get text in Bulk from Multiple PDF files

from handle_scanned_pdf import get_pdf_text_bulk_pdf

pdf_folder_path = 'pdf_files'
output_path = 'output'
draw_boxes = True
lang_code = ['en'] # 'eng'
ocr_used = 'easyocr' # 'tesseract'
lang_rtl = True
get_pdf_text_bulk_pdf(pdf_folder_path, output_path, lang_code, ocr_used, lang_rtl, draw_boxes)

Output:

{'number_of_files': 1,
 'txt_file_path_bulk': ['output/sample_.pdf'],
 'bounding_img_path': ['output/images_bounding/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_0.jpg',
  'output/images_bounding/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_1.jpg',
  'output/images_bounding/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_2.jpg',
  'output/images_bounding/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_3.jpg',
  'output/images_bounding/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_4.jpg']}

Get text from a single PDF file

from handle_scanned_pdf import get_pdf_text

pdf_path_ = 'pdf_files/sample_.pdf'
output_path = 'output'
draw_boxes = True
lang_code = ['ar', 'en'] # 'ara+eng'
ocr_used = 'easyocr' # 'tesseract'
lang_rtl = True
get_pdf_text(pdf_path_, output_path, lang_code, ocr_used, lang_rtl, draw_boxes)

Output:

{'bounding_img_path': ['output/images_bounding/pdf_bounding_images/text_with_boxes_pdf_0.jpg',
  'output/images_bounding/pdf_bounding_images/text_with_boxes_pdf_1.jpg',
  'output/images_bounding/pdf_bounding_images/text_with_boxes_pdf_2.jpg',
  'output/images_bounding/pdf_bounding_images/text_with_boxes_pdf_3.jpg',
  'output/images_bounding/pdf_bounding_images/text_with_boxes_pdf_4.jpg'],
 'txt_file_path': 'output/sample_.txt'}

Extract text, draw bounding boxes, and convert PDF file to text searchable PDF

from handle_scanned_pdf import scanned_pdf_to_text_searchable_pdf

file_pdf = 'sample_.pdf'
output_folder_path_img = 'img'
output_path = 'output'
lang_code = ['ar', 'en'] #'ara+eng'
image_converted_format = 'png'
ocr_used = 'easyocr' # 'tesseract'
ocr_used_txt_extraction = 'easyocr' # 'tesseract'
txt_extract_lang_code = ['ar', 'en'] #'ara+eng'
font_name = 'Scheherazade'
font_ttf_path = 'ScheherazadeNew-Regular.ttf'
font_size = 12
lang_rtl = True
non_standard_font = True
get_text=False
draw_boxes=False
scanned_pdf_to_text_searchable_pdf(file_pdf, output_folder_path_img, output_path, lang_code, ocr_used, ocr_used_txt_extraction, txt_extract_lang_code, font_name, font_ttf_path, font_size, lang_rtl, non_standard_font, image_converted_format, get_text, draw_boxes)

Output:

{'file_name': 'sample_',
 'img_path': 'output/img/sample__images',
 'pdf_path': 'output/searchable_pdf_sample_.pdf',
 'number_of_pages': 5,
 'text_file': {'bounding_img_path': [],
  'txt_file_path': 'output/sample_.txt'}}

Extract text, draw bounding boxes, and convert PDF file to text searchable PDF in Bulk

from handle_scanned_pdf import scanned_pdf_to_text_searchable_pdf_bulk

pdf_folder_path = 'pdf_files'
output_folder_path_img = 'img'
output_path = 'output'
lang_code = ['ar', 'en'] #'ara+eng'
image_converted_format = 'png'
ocr_used = 'easyocr' # 'tesseract'
ocr_used_txt_extraction = 'easyocr' # 'tesseract'
txt_extract_lang_code = ['ar', 'en'] #'ara+eng'
font_name = 'Scheherazade'
font_path = 'ScheherazadeNew-Regular.ttf'
font_size = 12
lang_rtl = True
non_standard_font = True
get_text=True
draw_boxes=False
scanned_pdf_to_text_searchable_pdf_bulk(pdf_folder_path, output_folder_path_img, output_path, lang_code, ocr_used, ocr_used_txt_extraction, txt_extract_lang_code, font_name, font_path, font_size, lang_rtl, non_standard_font, image_converted_format, get_text, draw_boxes)

Output:

{'number_files_converted': 1,
 'files_details': [{'file_name': 'sample_',
   'img_path': 'output/img/sample__images',
   'pdf_path': 'output/searchable_pdf_sample_.pdf',
   'number_of_pages': 5,
   'text_file': {'bounding_img_path': [],
    'txt_file_path': 'output/sample_.txt'}}]}

---

EasyOCR Searchable PDF Output Sample

---

References:

• pytesseract package
• tesseract-ocr
• Data Camp Pytesseract Guide
• Support Other Languages On Windows
• EasyOCR
• ReportLab