pyimgscan: open-source document scanning tool

Take a picture with your phone and convert it to a professional-looking scan.

* Usage

$ python3.9 ./pyimgscan.py

usage: pyimgscan.py [-h] -i IMAGE [-I [INVERTED]]

The program will by default output a corrected.png and a thresholded.png

• -h display the above help message
• -i specify a path to the source image file
• -I also output an inverted binary image (white on black) to thresholded_inverted.png

* Dependencies

numpy==1.16.2
opencv-contrib-python==4.2.0.32

can also be found at requirements.txt

* Setup with virtualenv

1. Set up virtual environment:

virtualenv -p python3.9 PyImgScan/
cd ./PyImgScan/
./bin/activate

2. Clone repository:

git clone https://gitlab.com/sammdu/pyimgscan.git

git clone https://github.com/sammdu/pyimgscan.git

3. Install dependencies:

cd pyimgscan/
pip install -r requirements.txt

4. Test out an example!

python ./pyimgscan.py -i ./test/1.png

Available examples: ./test/1.png ./test/2.jpg ./test/3.jpg

* TODO

• unify average brightness for better adaptation to different lighting of input images
• instead of convex hull, try to always complete a 4-corner-polygon
• try Gaussian thresholding for better binary image results
  https://docs.opencv.org/3.4.0/d7/d4d/tutorial_py_thresholding.html
• add shadow removal functionality
  https://stackoverflow.com/questions/44047819/increase-image-brightness-without-overflow/44054699#44054699
  https://stackoverflow.com/questions/44752240/how-to-remove-shadow-from-scanned-images-using-opencv
• add a border / four anchor points for better positioning and more reliable contour recognition
• build binaries with PyInstaller
• support multi-color thresholding

---

References:

imutils https://github.com/jrosebr1/imutils

PyImageSearch https://www.pyimagesearch.com/2014/09/01/build-kick-ass-mobile-document-scanner-just-5-minutes/