Comparison of methods for outlier identification in surface characterization

Raw data provided by measurement instruments like confocal microscopy often contains non-measured points and outliers. Ten statistical methods for outlier identification which can be implemented in the area of surface metrology are analysed and compared. These methods for outlier identification are introduced and their corresponding algorithms for data pre-processing before surface characterization are developed. Twenty-four Mat files were created based on two standard data sets provided by the National Institute of Standards and Technology. These files were assigned by four factors with two to three levels to represent all possible surface types. Based on processing the same series of contaminated data sets, the number of missed outliers, the difference of the height parameters, and the elapsed time by each method are compared. Algorithm efficiency, robustness, breakdown point, limitations, advantages, etc. are compared and analysed. Two of those ten methods were combined to know their potential. A type C1 spacing standard artefact was measured by 3D image confocal microscopy, and the data was processed by those algorithms. The difference of Sa and that of elapsed time are compared.