Ask Your Question

Revision history [back]

flannbasedMatcher trained with multiple sample


I have trained a flannbasedMatcher with 1 sample image(450 descriptors). The flannbasedMatcher initialize with default KDT for SURF/SIFT, LSH for ORB. I do knnmatch (k=2) with a test image (same image as the sample), I got ~400 match points.

I have trained a flannbasedMatcher with multiple sample (50 images each ~450 descriptors). Then I use knnmatch to match a test image (which is exactly one of the sample images). The most matched image is that sample image (expected as test image is same as one of the samples), but the match points are dropped to <40 points.

I have tried with ORB/SURF/SIFT, although number of descriptor is different, the result from matcher trained with multiple samples, the best matched image always has a much fewer match points than a matcher trained with a single sample. Match points is further decreased as the matcher trained with a larger number of samples.

I have also tried to trained a flannbasedMatcher with 2 exactly same images and use knnmatch. The result with the same image as query, is no match point found.

Is there any way to make a flannbasedMatcher trained with multiple samples, to give out similar number of match points as a matcher trained with single sample?