Correlated probability fusion for multiple class discrimination
Jackson O’Brien · 1999
Probability-level fusion often assumes independence of sources, in which case there are well established methods for combining probabilities (such as renormalised multiplication). However, in many real world data fusion applications the assumption of independence does not hold. In this case it is necessary to use more sophisticated algorithms which take into consideration the correlations present in the data. The fusion of correlation probabilities algorithm previously developed has been shown to allow superior performance over the renormalised multiplicative fusion of probabilities when using synthetic data from two correlated sources. This paper illustrates the technique using data from real air targets and extends the algorithm to deal with multiple classes.