Probabilistic Combination of Multiple Evidence

In pattern recognition systems, data fusion is an important issue and evidence theory is one such method that has been successful. Many researchers have proposed different rules for evidence theory, and recently, a variety of averaging rules emerged that are better than others. In these methods, the key issue becomes how to give the weights to the multiple contributing factors, in order to calculate the average. To get better weights for the multiple bodies of evidence, we propose the use of structural information of the evidence. The bodies of evidence lie on a certain informational structure which can be described by a probability distribution and the probability of each evidence can serve as a weight for the evidence. Our experimental results show that our method outperforms other previous methods. Key words: Evidence Theory, Data Fusion, Decision Making, Probability, Belief Function

• Type: Journal
• Year: 2009
• Where: ICONIP
• Authors: Heeyoul Choi, Anup Katake, Seungjin Choi, Yoonseop Kang, Yoonsuck Choe
• Keywords: Data Fusion, Evidence Theory, ICONIP 2009, Information Technology, Pattern Recognition Systems