How do I filter the false match returned from matchTemplate best match? Here are the sample images:
This is the sub image:
The sub-image is found correctly in this image:
But not in this image:
How do I check for a bad match (like the example above) and filter it out?
Example results:
The problem with the tutorial is that it knows the object is going to be in the image, and it does not cope with the fact that there are possible no matches at all. There are two main issues in this tutorial code if you want to apply it for your case.
normalize( result, result, 0, 1, NORM_MINMAX, -1, Mat() ); which means the smallest match (very low score) will still be normalized towards the 1 value.As you can see here:
/// Localizing the best match with minMaxLoc
double minVal; double maxVal; Point minLoc; Point maxLoc;
Point matchLoc;
minMaxLoc( result, &minVal, &maxVal, &minLoc, &maxLoc, Mat() );
Problem is that in your case, you will have to find a harsh threshold on the non-normalized match score map. If it does not meet the threshold, there will simply be no hit!
I solved the problem by using the NORMED variants and remove the normalize call, then use minVal or maxVal as similarity score. The good match returns > 0.99 and the bad match returns ~0.92, which is good enough for me since both differ only the text. You should update the example with this info and I can mark it as accepted.
Well we should not update the example, as it clearly shows that you take a part from an existing image and try to find it again using templates. However, feel free to add a second one with a more general case, for example find a logo in a promo video that tends to be gone from time to time.
Asked: 2018-05-08 04:59:53 -0600
Seen: 430 times
Last updated: May 08 '18
maybe show, how you call it, too ?
@berak the same code as the official tutorial, with modification to image paths: https://docs.opencv.org/2.4/doc/tutor...