Experiments on the Accuracy of Feature Extraction

Abstract

Feature extraction is an approach to visualization that ex- tracts important regions or objects of interest algorithmically from large data sets. In our feature extraction process, high-level attributes are cal- culated for the features, thus resulting in averaged quantitative measures. The usability of these measures depends on their robustness with noise and their dependency on parameters like the density of the grid that is used. In this paper experiments are described to investigate the accuracy and robustness of the feature extraction method. Synthetic data is gener- ated with prede ned features, this data is used in the feature extraction procedure, and the obtained attributes of the feature are compared to the input attributes. This has been done for several grid resolutions, for di erent noise levels, and with di erent feature extraction parameters. We present the results of the experiments, and also derive a number of guidelines for setting the extraction parameters.