SIFT Keypoint Location

I am currently working in SIFT , I created a difference in image levels of Gaussian and extrema. Can someone explain to me how to use the Hessian matrix to eliminate the small contrast key?

+6
source share
1 answer

A good key point is the angle. This comes from Harris' corner work and good tracking features (KLT), and then is highlighted by the publication of Mikolajczyk and Schmid .

Intuitively, the angle is a good function because it is the intersection of two lines, while one line segment can move along its direction, which leads to less accurate localization. A line segment is an edge, i.e. First order derivative (gradient). An angle is an edge that changes direction dramatically. This is measured by a second-order derivative, therefore, the Hessian matrix is โ€‹โ€‹used, containing the values โ€‹โ€‹of the second derivatives of the direction.

+1
source

Source: https://habr.com/ru/post/913821/


All Articles