Ask Your Question
1

cluster similar keypoints in a set of near similar images

asked 2016-02-01 08:34:49 -0600

501 - not implemented gravatar image

updated 2016-02-03 08:24:40 -0600

Hello, in a project I'm using SURF to extract keypoints from a set of near similar images. The images are near similar, so they have a lot of similar keypoints and a few keypoints that are different. My goal is to find groups of keypoints that matched by all images

is there an idea how to solve this? I can't use kmeans, because i doesn't know something about the k

UPDATE

My current approach is using a bruteforce-matcherto find the good matches of each image. I compare all images with each other. But it is very difficut to save the intersection from all good matches of all images. Is there a datastructure that can help me to solve this problem?

edit retag flag offensive close merge delete

Comments

Doesn't BOWTrainer helps you?

thdrksdfthmn gravatar imagethdrksdfthmn ( 2016-02-01 09:56:45 -0600 )edit

I think no, bechause I must set up the count of clusters

501 - not implemented gravatar image501 - not implemented ( 2016-02-02 02:15:31 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
2

answered 2016-02-01 17:41:03 -0600

Tetragramm gravatar image

Try RemoveDuplicated. You'll need to append the vectors together, but this should work, as long as the locations in the image are the same.

edit flag offensive delete link more

Comments

thanks!, that's a beginning for images, that shooted by a tripod. Any ideas how to handle keypoints that are near similar but shooted from a differet position?

501 - not implemented gravatar image501 - not implemented ( 2016-02-02 02:18:46 -0600 )edit

For that you would need to use the descriptor matchers and remove any good matches. I'm afraid I don't have enough experience with those methods to help.

Tetragramm gravatar imageTetragramm ( 2016-02-02 15:40:13 -0600 )edit

yes, that's theI the aproach I'am going. At the moment I compare all images with each other. But it is very difficut to save the intersection from all good matches of all images. Is there a datastructure that can help me to solve this problem?

501 - not implemented gravatar image501 - not implemented ( 2016-02-03 08:21:14 -0600 )edit

Well, you can either simply have a list of singular points. First image's points to begin with, then any points with no matches are appended to the list.

Alternatively, you can use the tree structure like in Connected Components. That is designed for holding intersections. You can see that inside the image processing module.

Tetragramm gravatar imageTetragramm ( 2016-02-03 18:31:56 -0600 )edit

Question Tools

2 followers

Stats

Asked: 2016-02-01 08:33:51 -0600

Seen: 874 times

Last updated: Feb 03 '16