deshade-doc

Remove shadow from hand-scanned documents image. My graduation researches at The Salesian Polytechnic.

Usage

Requirements

• Docker

Run

1. Prepare and put target image(s) into ./img/
   • images may hand-scanned documents.
   • if you want to evaluate the result, you also have to machine-scan your input, and put it with prefix scanner_
     • ex: note01.jpg and scanner_note01.jpg
2. Bootup docker, and bash
   • ./script/up.sh

• Pattern1: deshade-doc only
  • python src/source.py {img_name}
  • ex: python src/source.py note06.jpg
  • relative path from ./img/
• Pattern2: deshade-doc and evaluate
  • python src/evaluate.py {img_name}
  • not only evaluate. deshade-doc and evaluate.

Result

deshade-doc

• result/complete.png: complete image. not thresholded.
• result/lighted_up_shade.png: last processed shade's crop image.
• result/shade_cropped.png: shade before processed.
• result/thed.png: input image just thresholded with cv2.THRESH_OTSU
• result/thed_completed.png: thresholded, and completed image.

deshade-doc + evaluate

• result/thed.png, result/thed_completed.png: same above.
• result/thed_scanned.png: scanned image just thresholded with cv2.THRESH_OTSU for compare.
• result/scanned-m-base.png: comparing image via matching between ./result/thed_scanned.png and ./result/thed_base.png.
  • 
• result/scanned-m-deshaded.png: comparing image via matching between ./result/thed_scanned.png and ./result/thed_completed.png.
  • 
• STDOUT: matching distance as score. smaller is good(seems near with input image).
  • ex: note01.jpg (scanner_note01.jpg is already prepared.)

  •   result:
      base: 22.65706051873199
      deshaded: 23.282574568288855
    

  • by the way, result above is bad :p

Request, Question

Issue, mail is welcome.