matching surf features - degree of similarity -
i using opensurf find best matches in 2 images. finds matching points. wondering how can know degree of similarity between 2 matched points ((how strong match is). appreciate help. thanks.
this documented in literature, including surf paper itself. find distance (e.g. euclidean, mahalanobis) between descriptor vectors. since squared distance faster compute (it avoids square root), might see dot product of vectors used instead since equivalent squared euclidean distance.
standard practice decide whether or not accept match based on distance , threshold. sift paper (lowe 2004) gives more complicated way of accepting matches if recall correctly, might want read too.
in opensurf, descriptors float
vectors stored in ipoint
class - once have called surf.getdescriptors
, populated ipoint
vector given constructor, ipoint.desctiptor
fields of pair of ipoint
s , compute distance.
Comments
Post a Comment