Ask Your Question
0

OpenCV BOWKmeans and SVM

asked 2016-12-30 04:28:41 -0600

John Vision gravatar image

updated 2016-12-30 05:58:25 -0600

Hi,

I have question about using KMeans with SVM. I have implemented an algorithm that extracts desired ROI from video frame with %80 accuracy. The original paper use geometrical approach to decide whether candidate region is a desired region or not desired one. In my school project, I want to improve that geometrical refinment part with local feature based refinment methods. Therefore I want to use both unsupervised and supervised learning methods on extracted ROI to see that it can be improved or not by local features. My desired output is to give an answer to is that desired ROI or not with higher accuracy.

So I setup a dictionary with BowKMeans. But I could not figure out how to combine it with SVM in OpenCV. Can anyone guide me how to do that?

My code is adapted from here and I also eximined this link but could not understand how to do.

edit retag flag offensive close merge delete

Comments

what are you trying to achieve with it ?

you need labels for any kind of ml classification, finding nearest neighbours (e.g. using a flann::Index) would be a different story.

berak gravatar imageberak ( 2016-12-30 04:44:39 -0600 )edit

Hi, Thank you for reply. What I'm trying to achive is using unsupervised learning output, which is my clusters from BowKMeansTrainer, with SVM. If I cannot use this approach, can you suggest me a methodsto test my dictionary with an image. Thank you again.

John Vision gravatar imageJohn Vision ( 2016-12-30 05:21:59 -0600 )edit

let me ask again, what is the goal of your unsupervised learning, what is the desired output ?

would it be: "finding similar images" ?

berak gravatar imageberak ( 2016-12-30 05:42:52 -0600 )edit

Thank you again for your reply. To be more clear: I have implemented an algorithm that extracts desired ROI from video frame with %80 accuracy. The original paper use geometrical approach to decide whether candidate region is a desired region or not desired one. In my school project, I want to improve that geometrical refinment part with local feature based refinment methods. Therefore I want to use both unsupervised and supervised learning methods on extracted ROI to see that it can be improved or not by local features. My desired output is to give an answer to is that desired ROI or not with higher accuracy. Now I'm editing my question to be more clear about it. Thank you again break

John Vision gravatar imageJohn Vision ( 2016-12-30 05:56:20 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
0

answered 2016-12-30 05:59:56 -0600

John Vision gravatar image

Yes it is definitely finding similar images.

edit flag offensive delete link more

Question Tools

1 follower

Stats

Asked: 2016-12-30 04:28:41 -0600

Seen: 289 times

Last updated: Dec 30 '16