Information Fusion in Signal and Image Processing: Major Probabilistic and Non-Probabilistic Numerical Approaches

Chapter 12

Conclusion

Now that we have been through this overview of the major numerical fusion methods and of their use in signal and image processing and in robotics, we will sum up a few achievements and conclusions, as well as a few issues that still pose difficulties.

12.1. A few achievements

The previous chapters have shown that a wide variety of numerical techniques are used for the fusion of imprecise and uncertain information. This diversity is the result of the diversity of the tasks themselves that contribute to the decision in a multi-source information system. Probabilistic methods remain the most commonly used, mostly because they have led to the development of operational tools and great know-how, which is the result of a considerable amount of practice. On the decision level these tools turn out to be particularly efficient, whereas for modeling, some aspects remain limited or even disputed. Fuzzy set theory relies on a type of modeling close to intuition. In the fusion applications mentioned here, there is still little formalism or development in the decision phase. On the other hand, the combination phase is very rich and allows knowledge of any type to be included. Belief function theory offers the most powerful modeling tools, making it possible to simply and efficiently include knowledge, imprecision and uncertainty. Combination, as it is used in signal and image processing, is limited to the conjunctive mode.

In numerical fusion for signal and image processing, ...