A study on the fast M-estimation based Gaussian filter controllable in robustness

Abstract

A FMGF (Fast M-Estimation based Gaussian Filter) is the fast robust Gaussian filter that output wave pattern accorded mutually with the Gaussian filter output.. However, in some case, robustness of FMGF is less than Brinkmann's robust Gaussian filter. Then, robustness of the FMGF is strengthened. The FMGF has characteristic that robustness and output process is calculated in each x-coordinates. Using this characteristic, only few x-coordinate that is affected by outlier are recalculated. In the recalculation process, parameter m value changes to small value, and to strengthens robustness. Using characteristic of the FMGF that accords with output of the Gaussian filter, the outlier areas are judged. The first process of the new FMGF, executes the Gaussian filter and normal FMGF. Next, only x-coordinates that has deviation between output of the Gaussian filter and the FMGF are recalculated. It is confirmed that the robustness of the new FMGF is strengthened without problems.