# Why do we use Hough instead of RANSAC in SIFT?

In his SIFT paper, why did Lowe choose to use a Hough transform rather than RANSAC to recognize clusters of 3 consistent features? (Note that RANSAC is more efficient in comparison with Hough)