Parameter setting support for a 3D images processing systems
Résumé
This paper is concerned to a fusion system devoted to 3D images interpretation. This task consists in identifying typical areas within the images to understand a complex phenomenon. Classical approaches consist of measuring the detection quality by means of a numerical indicator and then to find a set of parameters that increase the chosen indicator. End-users are able to express his satisfaction on an obtained detection. This satisfaction encompasses all the subjective aspects that are inherent to this kind of the evaluation. Then an approximate model linking the end-users satisfaction to the set of parameters could be reach. Such model is then used to compute a predictive satisfaction and the only a priori satisfactory detection can be tested. So each new predictive satisfaction must be compared with the previous ones to be sorted or ranked. The parameter adjustment problem can be seen as a decision problem. The objective is to avoid the computation of many fusions, because of the entire time consuming process. This paper addresses the MCDA and the multivariate analysis approaches to link the decision data (observations, attributes) with the satisfaction information in order to identify an interesting model linking the parameters to the detection.