EasyOCR - 表格提取

Question

I use easyocr to extract table from a photo or scanned PDF, but I have a problem in fine tuning the data as a table. I try to make a searchable pdf according to extracted coordinates but when I convert it to csv, the lines are not tune. I would appreciate if someone guide me about this.

Answer 1

You can use my package for this : ocr-nanonets-wrapper. It works for both images and pdf.

Download the package using pip: pip install ocr-nanonets-wrapper

Get an API Key. This key is free and gives you unlimited access to use the package.

• Go to nanonets.com and signup
• On your Nanonets Account, Go to My Account -> API Keys
• Copy your API Key

Sharing code below to get tables as csv -

from nanonets import NANONETSOCR
nanonets = NANONETSOCR()

nanonets.set_token('YOUR_API_KEY')

nanonets.image_to_csv('INPUT_FILE_PATH', filename = 'OUTPUT_FILE_NAME.csv')
nanonets.pdf_to_csv('INPUT_FILE_PATH', filename = 'OUTPUT_FILE_NAME.csv')

You can leave filename blank as well, that will just take your file name and append ".csv" to it. The .csv output file will be created in your current directory.

Hope this helps :)

Answer 2

As far as I know, easyocr currently does not support table recognition. The best table recognition should be PaddleOCR's PP-Structure model. This is what I use now, and the effect is very good. You can try it.

link: https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/README.md