Boosting the semantics sensitive satellite image retrieval using a voting algorithm

This paper proposes a semantic inference approach, which utilizes a group of context-sensitive Bayesian networks to infer the semantic concepts based on different regional spatial relationships, i.e. disjoined, bordering, invaded by, surrounded by, near, far, right, left, above and below. Each Bayesian network performs the inference based on one kind of the regional spatial relationships. Finally, a voting algorithm is proposed to combine the group of the Bayesian networks into a more accurate and robust semantic concept classifier. The experiments using IKONOS imagery show that the precision of the proposed voting algorithm is consistently higher than that of the single context-sensitive Bayesian network.