A comparative experimental study of spectral hashing

Abstract

Binary encoding methods that keep similarity in large scale data become very used for fast retrieval and effective storage. There have been many recent hashing technics that produce semantic binary codes. We are particularly interested in Spectral Hashing based methods which provide an efficient binary hash codes in a very simple way. This paper presents a comparative experimental study of Spectral Hashing to show the performance gain and the behaviour of this method on large scale Databases. In the best of our knowledge there is no experiments done on the evolution of the hamming matrix size on big data. Two large databases are used to show the limitation of Spectral Hashing and possible research tricks will be proposed.