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-pdfServer 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.
- 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-ocrOnly if required set the below path to Tesseract executable:
pytesseract.pytesseract.tesseract_cmd = r'/usr/bin/tesseract/tesseract.exe'- Does not require server installation.
- Supports reading more than 1 language at a time.
- Performes faster on a GPU.
- List of supported language code.
- 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"Defining the language codes for EasyOCR:
lang_code = ["en","ar"]
txt_extract_lang_code = ["en","ar"]pytesseract===0.3.10
pdf2image===1.17.0
PyPDF2===3.0.1
opencv-python- Straightforward functions.
- Customizable process.
- JSON output.
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/images_bounding/3ba4c1f1-775f-4e05-ab48-a40617087a57-1_bounding_images/text_with_boxes_3ba4c1f1-775f-4e05-ab48-a40617087a57-1_0.jpgfrom 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){'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']}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){'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'}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){'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'}}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){'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'}}]}