A Robust Gaussian Filter Corresponding to the Transmisson Characterisic of the Gaussian Filter

Abstract

A surface roughness profile of an object can be measured by extracting a mean line of the long wavelength component from the primary profile, and by subtracting it from the primary profile. This mean line is usually computed by convolving the traditional Gaussian filter (GF) and the primary profile. However, if an outlier exists in the primary profile, the output of a Gaussian filter will be greatly affected by the outlier. To solve the outlier problem, several schemes of robust Gaussian filter have been proposed. However there are several fatal problems that a mean line determined with respect to the measurement data containing no outliers does not agree with the mean line of the Gaussian filter output. To solve these problems, this paper proposes a new robust Gaussian filter based on a fast M-estimation method (FMGF) and the performance of the new robust Gaussian filter was experimentally clarified. As a result, if an outlier exist, the proposed method behaves a robust performance. If no outlier exists, the output wave pattern, RMSE and transmission characteristic accorded mutually with Gaussian filter.