A Hybrid Method for Relevance Feedback in Image Retrieval Using Rough Sets and Neural Networks

Image retrieval system is developing driven by the explosive growing of multimedia data. However the oneshot result can’t fit the human perception perfection. Relevance feedback is utilized to improve the retrieval performance. In this paper, a new relevance feedback algorithm-ARFRS is proposed. This algorithm is composed of two parts: pre-processing part and classification part. The pre-processing part use Rough set theory to induce the classification rule and reduct set of the image features. The classification part uses neural network ensembles to train the retrieval return images based on the classification rules and reduct set. The algorithm makes full use of the advantage of rough set theory and neural network ensembles to represent the human perception objectively and improve the retrieval performance. Through the experiments, we demonstrate that the retrieval performance is improved significantly by using our algorithm. Copyright c © 2003-2005 Yang’s Scientific Research Institute, LLC. All rights reserved.