| 主权项 |
1. A computer-implemented method, implemented by hardware in combination with software, the method comprising:
for each particular template image of a plurality of template images: (A) determining a set of feature points associated with a template image; (B) for each particular test image of a plurality of test images:
(B)(1) attempting to match a particular set of feature points associated said particular test image with the set of feature points associated with the template image, said attempting to match producing a match set comprising pairs of feature points, each pair containing a feature point from the template image and a corresponding matching feature point form the particular test image; (C) for at least some test images of said plurality of test images: evaluating said attempting to match in (B) to determine
(i) a first set of true positive matches in said test images of said at least a portion of said template image in said particular test image, and(ii) a second set of false positive matches in said test images in which a portion of said template image is not considered to be in said particular test image; (D) training a match-verification classifier for said template, said training using (i) said first set of true positive matches, and (ii) said second set of false positive matches, wherein an input vector to the match-verification classifier includes at least one element of information for each of the feature points for the template, wherein at least one element of information for each point includes at least a presence of the feature point in the match set; and (E) associating said match-verification classifier with said template image. |
