PROJECT TITLE:

EMR A Scalable Graph-based Ranking Model for Content-based Image Retrieval - 2015

ABSTRACT:

Graph-primarily based ranking models have been widely applied in info retrieval area. In this paper, we tend to target a well-known graph-based model - the Ranking on Information Manifold model, or Manifold Ranking (MR). Particularly, it's been successfully applied to content-based image retrieval, as a result of of its outstanding ability to get underlying geometrical structure of the given image database. However, manifold ranking is computationally terribly expensive, that considerably limits its applicability to giant databases especially for the cases that the queries are out of the database (new samples). We propose a completely unique scalable graph-primarily based ranking model referred to as Efficient Manifold Ranking (EMR), making an attempt to deal with the shortcomings of MR from 2 main views: scalable graph construction and efficient ranking computation. Specifically, we tend to build an anchor graph on the database instead of a ancient k-nearest neighbor graph, and design a replacement type of adjacency matrix used to speed up the ranking. An approximate method is adopted for economical out-of-sample retrieval. Experimental results on some massive scale image databases demonstrate that EMR may be a promising technique for universe retrieval applications.