Ask Your Question
0

template matching invariant to rotations and noise

asked 2019-10-02 09:29:04 -0500

IS any way to enhance the maximum value of similarity in case the template on image is rotated with different angle than the saved template

The problem here is in case there is stamp on image rotated with small angle the template matching returns maximum matching value with small value so the program cannot now if this value is due to rotation or noise or due to the stamp is not found in the image

edit retag flag offensive close merge delete

Comments

Did you try using invariant descriptors to rotation like (SIFT (sift is patented for commercial use) ,SURF,ORB,...), i think it's the best method if you are trying to find the rotation of your image compared to a reference image.

Kitnos gravatar imageKitnos ( 2019-10-02 09:49:21 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
-1

answered 2019-10-02 09:49:11 -0500

Czak gravatar image

There is no algorithm that is truly invariant to rotation of template, and to noise some algorithms may be more robust (i would advice zero normalised cross correlation with can be efficiently implemented using fft). And multiple minima can be avoided by cropping image before and find minima in croped sections, it will be helpfull if You Can make aditional test on localisation that would differenciate between minimas.

edit flag offensive delete link more

Comments

4

Not exactly there is Fourier Mellin transform

LBerger gravatar imageLBerger ( 2019-10-02 13:53:40 -0500 )edit

I find some good example of this algorithm. in practice, I found that Fourier based template matching works not so well if there are some other patterns on the be-matched images.

I think the rotation invariant matching or object detection is important topics for the application of computer vision in real world but with very few spot lights.

gino0717 gravatar imagegino0717 ( 2019-10-04 02:20:53 -0500 )edit
1

It's hard to give any advice without seeing data. If you can post images that you want to match it will we easier. My expertice comes from reaserch on digital image correlation where there is loads of repetetive paterns and shear, rotation and non linear deformation, but scale remains same, and FFT base metods are very common in that field. Mayby resolution of template do not match resolution in image that you try to match? Mayby scale it to diffrent sizes and look for the results.

Czak gravatar imageCzak ( 2019-10-04 03:19:54 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2019-10-02 09:29:04 -0500

Seen: 49 times

Last updated: Oct 02