anuraggupta29/document-scanner

Desktop based GUI Document Scanner with OCR

Document Scanner and OCR

For GUI version of this project go to : https://github.com/anuraggupta29/DocScan-electronJS

In Progress

• This is a desktop based GUI Document Scanner.
• It can scan multiple images at once.
• It automatically detects the 'document' part from the image.
• It corrects its orientation and perspective.
• The GUI will have the feature to manually select the 'document' part from each image, in case there's an error in automatic detection.
• It can save the image in multiple modes and resolution.
• It can also save all the images as a simgle pdf.
• It will also have the feature of optical character recognition.

How document detection works on each image?

• It resizes a copy of the image for manipulation.
• It then grayscales the image.
• Then it applies a bilateral filter to it (blurs the image while preserving edges).
• It produces a canny image (binary edges only) with the threshold dependent on median of intensity values of image pixels.
• It then detects the contours in the image (the curves).
• Then it takes the largest contour based on perimeter (As the document part will have the largest perimeter.
  Note : The document will also have the largest area but most of the time the contour detected is not a closed curve, because of which its area becomes very small but the perimeter remains large.
• Next it takes a convex hull of our selected contour (smallest polygon enclosing the contour).
• Then it uses approximatePolyDP to approximate the convex hull as a rectangle.
• Thus it obtains the 4 corners of our document.
• Then it scales to coordinates according to the original image size.
• From those 4 coordinates we identify which one is the tl, tr, br, bl coordinate.
• The it calculates the height and width of our document portion.
• Then it does a perspective transform of the original image with our rectangle coordinates and produce the transformed image.

Note : Rest of the features are yet to be added.