Can I find out how the training data matches the current data?

asked 2017-12-14 01:26:27 -0600

I am doing character recognition through knn using opencv 3.2.0

When recognizing, two result come out. If k is 1, how much does the training data match the current data as a percentage?

edit retag flag offensive close merge delete

Comments

can you show, what you're doing (aka: code) . also, try to read the tutorials herel.

language used ?

berak ( 2017-12-14 01:33:22 -0600 )edit

I recognized some chars through open source found on the web. https://github.com/MicrocontrollersAn...

and I made new training data and then recognizing
but There are many similar characters in Korean like ㅁ ㅂ so I want to know the knn result accuracy

borocoli ( 2017-12-14 22:52:50 -0600 )edit

add a comment

1 answer

Sort by » oldest newest most voted

answered 2017-12-14 04:03:08 -0600

kbarni

4134 ●1 ●13 ●49

You can create a validation base; i.e. a set of character images from your dataset with the label (like image of an "A" with the label A). Labelizing 100 images won't take long.

You run these images through the classifier, and you check how many correct results do you get. It will give the precision of the classifier. You can change the parameters (like K) to see how does it behave.

If you can't improve the results, you might have to use a different training set or another classification algorithm.