matching keypoints

• related: SIFT - local detector

• given two keypoints, we can calculate their Histogram of Oriented Gradients (HoG) features

• can calculate matching score $\sqrt{{\Sigma }_i(\text{HoG}(k_1{)}_i-\text{HoG}(k_2{)}_i{)}^2}$

  • find nearest neighbor for each keypoint in image A in image B

• sensitive to number of histogram orientations?

• stability to noise

  • match features after random change in scale + orientation w differing noise levels
  • find nearest neighbor

• stability to affine changes