I'm solving image matching task using SIFT as a feature extractor
To match obtained descriptors I used FLANN based Matcher (provided by OpenCV) and result looks good:
But applying same algorithm to different images I still have some "False Positive" matches because of shapes similarity:
I understand, that tuning hyperparameters can do great things here, but look a little closer to wrongly matched key points Most of their connecting lines cross other ones, which means that the mutual arrangement of matched points on source image (Pikachu) and mutual arrangement of matched points on target image (website screenshot) are different
FLANN based Matcher (as well as all other matchers I found) doesn't pay attention to key points placement. It match only their descriptors
Are there algorithms that can match SIFT's key points of 2 images using their mutual arrangement as well as descriptors?
