A combination of graph ranking and term collocation information for image annotation

Abstract

In a large image database, an automatic image annotation plays an important role to assign a caption to an image. There are many applications of image annotation which include information extraction, knowledge acquisition, finding answers from specific questions, etc. In this research, we propose an image annotation approach using a graph ranking algorithm on a novel structure which includes similarities among regions within images and caption term collocation information. Caption terms are selected using a graph random walk method biased toward the query image. A series of experiments are performed on standard image data sets to evaluate the proposed approach. The results show the annotation accuracy of 56.96%.