DOG DETECTOR

This detector relies on difference of Gaussian (DOG) in order to find keypoints

$$DoG(x,y,\sigma )=(G(x,y,k\sigma )-G(x,y,\sigma ))\ast I(x,y)=L(x,y,k\sigma )-L(x,y,\sigma )$$

This approach is more computational efficient of computing LOG, and it’s a good approximation of the results

DOG COMPUTING

Computation of dog is done by down-sampling and gaussian smoothing the input image in order to obtain the scale space and then by computing differences between adjacent scale levels

The next scale level is computed by taking half of the columns and rows and computing the same filter (performance optimization)

a point is detected as a feature if it’s DoG is an extreme of its 26 neighbors

DOG IMPROVEMENTS WITH FILTERS

In order to localize keypoints in an accurate way and remove unstable point filter procedures are needed

PREVIOUS NEXT