Best method to recognize currency notes

asked 2016-12-13 23:56:39 -0500

Hi,

I am quite new to OpenCV therefore excuse me if I ask silly questions.

I am working on an android project to recognize currency notes to help blind people. I have been using Cascade Classifiers.

Link1 Link 2

It did not give me good results ( maybe something wrong with the negative samples I used )

Then I used ORB algorithm. Here I managed to do a feature matching and recognize the currency note, the issue is there are very high number of false positives. I also read I could use SURF( I did not try it yet). This is a research project, therefore I think I can use SURF but it is also an improved version of SURF and it uses just one image for the matching. That way there could be multiple errors compared to a model trained.

If someone has done this successfully before please let me know, I would like to know the best way to do this and pivot my research and implementation to one method.

edit retag flag offensive close merge delete

Comments

1

none of your attempts will lead to something reliable there. (cascade classifiers can only detect a single object type, and feature matching is not meant to be used for classification)

you probably should try to seperate the bill detection from the classification problem, like:

  • find contours, check for rectangles in appropriate size
  • crop and align to xy axes
  • use machine learning on cropped images to classify
berak gravatar imageberak ( 2016-12-15 02:28:03 -0500 )edit

I am sorry for the delayed reply, somehow I did not get a notification on this reply to my mail. :(

Using machine learning independently? like using image's other features and trying to build a model?

akshika47 gravatar imageakshika47 ( 2016-12-22 23:11:19 -0500 )edit

Did you have any success? i am doing the same project for the blind people in my country so if you could share the code i will appreciate it very much

Mehrdad73 gravatar imageMehrdad73 ( 2019-01-17 12:50:04 -0500 )edit