Relevance Feedback Based on Particle Swarm Optimize Weight-Vector for Image Retrieval

Abstract

As the physical meaning of components are different in feature vector, this paper presents a weight query vector based on QPM to represent the user’s true intention more properly, and then proposes two RF frameworks to learn the weights for positives and negatives in the feedback process of CBIR by PSO. Experiments were conducted to validate the proposed frameworks based on color histogram weight-vector. The proposed frameworks were compared and outperformed four other relevance feedback methods regarding their efficiency and effectiveness, thanks to the fact that they can make full use of the features’ different importance.