-1

Augmented Reality supports basic text recognition through the Vuforia library and it supports the only computer generated fonts, I would like to do more by extending the functionality to the humanly written fonts. I just want to start it initially from numbers and later try with alphabets. Please help by sharing your thoughts on the feasible options available at this time?

Use case description: I want to build an AR app which will recognize the numbers written on a paper(assuming individual digits) for example if I show the paper which has 1 on it shall be recognized by the app and numerical 1 will be superimposed on the paper.

Suresh Kumar Tathari
  • 28
  • 1
  • 5

1 Answers1

0

This questions is very broad so I will give you a broad answer. What youre looking for is a python API which uses a model trained on MNIST dataset. This application should or could use Hough transform, homography and SIFT features. Use contour detection to split individual alphabets. The contour detection might or might not work given a how close the alphabets are to each other and if the handwritten text is in boxes or continuous. Here are a few links to get started with.

https://machinelearningmastery.com/handwritten-digit-recognition-using-convolutional-neural-networks-python-keras/ Training of a ML model

https://www.learnopencv.com/image-alignment-feature-based-using-opencv-c-python/ Image registration

Again , this question is not relevant to SO. Reddit might be a better choice.

Doomski
  • 117
  • 1
  • 11