Overview
Our goal is to learn image similarity models. We define
the similarity of two images P and Q according to their
squared Euclidean distance in the image embedding space:
D(f(P ), f(Q)) = kf(P ) − f(Q)k2 2 (1)
where f(.) is the image embedding function that maps an
image to a point in an Euclidean space, and D(., .) is the
squared Euclidean distance in this space. The smaller the
distance D(P, Q) is, the more similar the two images P and
Q are. This definition formulates the similar image ranking
problem as nearest neighbor search problem in Euclidean
space, which can be efficiently solved via approximate nearest neighbor search algorithms.