Robustness of the Method

Various tests were carried out to prove the reliability of the proposed methodology. A large number of test contours with different deviations from ideal shape and different missing parts were prepared in order to see how well the algorithm can perform in varying conditions.

The robustness of the algorithm is illustrated in figures 8 and 9. Figure 8 shows that our method is more robust than both the original HD and the method that minimizes the mean Hausdorff distance between the two contours. The original HD that minimizes the maximum distance results in a wrong matching. Minimizing the mean HD is closely related to the conventional least squares regression. This method also `smears' the defect.

  

Figure 8: Upper row: Our method. Lower row: The original and the mean HD.

Figure 9 demonstrates that our method can cope with incomplete and noisy measurements. Despite these factors, a good result is obtained.

  

Figure 9: An incomplete measured contour and its matching.