enter image description here

  • Above image has handwritten hindi / deavanagari letters all on a single

  • Task is to Prepare a dataset from the given assignment.jpg image using image

    processing tools
  • Idon't have clarity whether i can use photoshop editing tools like adobe photoshop etc.

  • If image processing tools mean opencv , is it really possible to create like a emnist like dataset from single image.

  • I am not able to come up with any way to seperate the letters from eachother

  • 1
    $\begingroup$ You have non overlapping crystal clear bitmaps, there is no difficulty in extracting the bounding box of each letter. What do the letters (foreground) have, that the background (paper) does not? What do the pixels that belong to the same letter have in relation to the pixels that belong to a different letter? $\endgroup$
    – A_A
    Commented Mar 27, 2019 at 8:57
  • $\begingroup$ @A_A , i think foreground letters have color black and background is white.I don't know how to find relation to pixels that belong to different letter.Even if it is possible i think it involves image processing, I don't know any concepts in image processing. $\endgroup$ Commented Mar 27, 2019 at 16:48
  • 1
    $\begingroup$ These characters are almost segmented already :) So how about defining a fixed grid on top of the image and cropping out each cell? If the characters in a next image (presumably you would have multiple such sheets) don't move a lot, then you can always segment them with this simple way. $\endgroup$ Commented Apr 27, 2019 at 7:43

1 Answer 1


Presumably this is an academic assignment of some sort and you are expected to use a specific tool, OpenCV I am guessing. You should review the documentation for that tool and try to find a feature which matches the requirements. I suspect you would need to use the Image Segmentation feature: https://docs.opencv.org/3.4.3/d3/db4/tutorial_py_watershed.html


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.