Application of quantitative MCDA methods for parameter setting support of an image processing system

This paper proposes to use quantitative methods to identify a preference model reflecting the overall satisfaction of the user according to the numerous parameters of a complex fusion system. The studied fusion system is devoted to 3D image interpretation and it works in interaction with experts who have knowledge and experience of the concerned applications. Such a system involves many sub-parts and each of them has many parameters that must be adjusted to obtain interesting detections. The link between the parameters and the overall satisfaction expressed by the experts is a priori unknown and it is a key issue to better interact with the system. After the presentation of the preference model relevance with the problematic, three model identifications (multivariate, UTA+ and MACBETH) are attempted in this paper to find an interesting set of parameters according to the available overall satisfaction. Obtained results show the complexity of this kind of identification, mainly because of the non monotonicity of the parameter utilities.