jupyter-anywidget-tesseract-pdfjs 0.0.4

No project description provided

jupyter_anywidget_tesseract_pdfjs

Tesseract./ pdf.js anywidget for previewing PDF and extracting text from PDF, image, etc. in JupyterLab

Inspired by and building on @simonw's (Simon Willison) OCR tool [about], use tesseract.js in a Jupyter notebook environment via an anywidget wrapper.

Using the anywidget framework, we can essentially load Javascript and WASM models into a sidebar widget and use the widget for "side-processing" using the browser machinery.

For example, we can use the tesseract.js for OCR/text extraction on images, and pdf.js for converting PDF documents to images which can then be OCR'd using tesseract.js.

This reduces the number of Python dependencies that need to be installed on the host machine, albeit at the expense of loading resources into the browser.

I'm not much a packaging expert, so some assets are likely to be loaded from a URI; ideally, everything would be bundled into the anywidget extension.

Installation

pip install jupyter_anywidget_tesseract_pdfjs

Usage

Import the jupyter_anywidget_tesseract_pdfjs package and launch a widget:

from jupyter_anywidget_tesseract_pdfjs import tesseract_panel

t = tesseract_panel()
#t = tesseract_panel("example panel title)
#t = tesseract_panel(None, "split-bottom")

This loads the widget by default into a new panel using jupyterlab_sidecar.

You can then drag and drop an image file or PDF file onto the landing area or load an image or path in from a notebook code cell.

Filetype	Local file	Web URL
Image	File drag / select; widget.set_datauri(?)	widget.url=?, widget.set_url(?), widget.set_datauri(?)
PDF	File drag / select	widget.pdf=?, widget.set_url(?)
Image Data URI	widget.datauri=?	N/A

Accessing extracted text

We can access extracted text via: t.pagedata

The results object takes the form:

{'typ': 'pdf',
 'pages': 3,
 'name': 'sample-3pp.pdf',
 'p3': 'elementum. Morbi in ipsum sit ...',
 'processed': 2,
 'p1': "Created for testing ...”, knowing'
}

The keys of the form pN are page numbers; the processed item keeps a count of pages that have been processed; the pages item is the total number of pages submutted for processing.

We can also review the extracted text for the last processed image: t.extracted

Review a history of files that have been processed: t.history

Exanples

Example code. Also the notebooks in examples:

# Image at URL
image_url = "https://tesseract.projectnaptha.com/img/eng_bw.png"
t.set_datauri(image_url)

#New cell
# We also need to "manually" wait for processing to finish
# before trying to imspect the retrieved data
t.pagedata



# View history of OCR lookups
t.history