svidovich/pdf-vision

( Try to ) Read PDFs with computer vision

pdf-vision

I'm starting this project because I have a book I want textified that's in PDF format, but it appears to just be scanned pages.

Structure

You can have a scratch/ folder to stash your items, and the gitignore will handle that.

Dependencies

We require:

• attrs for data classes.
• PyPDF2 for reading PDFs.
• Pillow for handling images.
• Tesseract for OCR.
• OpenCV for image pre-processing.

For my application, I installed the Serbian language pack.

apt install tesseract-ocr-srp

Getting Started

We recommend a virtual environment.

$ python3 -m venv venv
$ source venv/bin/activate
$ python3 -m pip install -r requirements.txt

Sample Input

The book I used to test this program is a Serbian historical text, and can be downloaded from Google Drive: https://drive.google.com/file/d/1ViI2Hq5ohhPO1pM-u_i2Lv3MAUrYS-l4/view?usp=sharing

Sample Output

Input Image:

Text Output:

Међу митским бићима у која је веровао 'а мести- мично и данас верује сеоски народ источне Србије и Баната, својом занимљивошћу и архаичношћу истИ- че се женски шумски демон — шумска мајка.

The entire output text ( unadulterated, I haven't fixed it up at all ) can be downloaded as a tarchive here: https://drive.google.com/file/d/1eLnLqTay_zmiPFjzPOuG1D3fOoMJMRCi/view?usp=sharing