Efficient Implicit Content-based Image Re-ranking Approach

Abstract

This paper presents a new image re-ranking approach that can implicitly improve the retrieved images based on the file’s contents and some user-specific actions. In more detail, multiple descriptors are used to describe image files accurately and they do not require user intervention or tuned parameters. Furthermore, each of these descriptors has a weight, which affects the file rank. Unlike existing approaches, descriptor weight is assigned dynamically and changes from one file to another based on the percentage of differences found by the descriptor. Hence, the developed weight mechanism improves the chance of getting the required files significantly. The performance of the developed approach is investigated through several experiments and it has been observed that the approach has the ability of showing the most relevant files at the top of the query results and increases the percentage of the retrieved relevant files.