Conclusion

We showed that substantial gains in performance in image interpretation can be achieved by fusing the opinions of multiple experts. We developed a theoretical framework for the combination of soft decisions generated by experts employing distinct object representations. By taking the confidence of the individual experts into account we derived weighted benevolent fusion strategies. One of the strategies was applied to the problem of automatic detection of microcalcifications in digital mammograms. The results demonstrated that the use of weighting factors leads to superior performance as compared with an equal weighting of experts opinions. Although the performance of the equally weighted combiner is better than the worst expert, only weighted fusion is always likely to outperform the best expert.

The experimental results also warn about a possible degradation in performance caused by the inclusion of a poorly performing expert. The use of weighting factors in multiple expert fusion provides a practical way of preventing such an event.