which are revisited latter in
Section Star Student Project . In this paper, we argue that
for scalable visual search, directly applying previous distributed
Star Student Project information retrieval techniques is
unsuitable. This argument will Star Student Project be
qualitatively explained in Section III and quantitatively Star
Student Project validated in Section Star Student Project
. Different from traditional distribution paradigms, we employ visual
feature statistics, such as visual Star Student Project word
concurrence and redundancy, for Star Student Project an
optimal distribution strategy. Our goal is to significantly reduce
the computational Star Student Project load via a distributed
visual search architecture, so that massive reference images say
millions or billions Star Student Project can be searched
based on a group of laptops or regular servers, or on cloud Star
Student Project computing
resources.