Evidence Combination Using Likelihood Generators
David B. Sher · 1987
Here, I address the problem of combining output of several detectors for the same feature of an image. I show that if the detectors return likelihoods I can robustly combine their outputs. The combination has the advantages that: * The confidences of the operators in their own reports are taken into account. Hence if an operator is confident about the situation and the others are not then the reports of the confident operator dominates the decision process. * A priori confidences in the different operators can be taken into account * The work to combine 'N' operators is linear in 'N'. This theory has been applied to the problem of boundary detection. Results from these tests are presented here.