Outlier filtering: a new method for improving the quality of surface measurements

TLDR: In this article, the authors proposed a new filtering technique that aims to improve the quality of measured surface data by removing measurement artefacts, such as spikes and batwings, that impact the data analysis.

Abstract: This paper introduces a new filtering technique that aims to improve the quality of measured surface data by removing measurement artefacts, such as spikes and batwings, that impact the data analysis. The methods currently available for measuring surfaces allow a large number of heights on a surface to be measured with high lateral and vertical resolution. However, measured data often contain outliers, which take the form of sharp peaks on the surface (Dirac type) and are particularly common in optical measuring methods. By nature, these peaks cannot belong to a 'real' surface; thus, an outlier filtering step is necessary and should be the initial preparation step for the measured data prior to any further analysis. The measurement artefacts can strongly influence the topographic characterization parameters and adversely influence quality control efforts as well as functional analyses for discrimination and correlation of the surfaces. The proposed method makes the filtering of such outliers easier and more effective with criteria linked to the standard deviation (Peirce method) and associated with a modal form-filtering method that is independent of the presence of these peaks. The filter is then applied to the surface at different scales; thus, the method is scale sensitive and improves the filter efficiency. This new method is applied to surface data that were measured by optical means from different surfaces. The surfaces were measured with a 3D-measuring laser microscope and laser scanning confocal microscopy. The results of testing these examples will determine the extent to which this method can improve the quality of measured data and thus influence the results of further analyses.