Virtual Image Correlation uncertainty

TL;DR

The paper introduces an improved Virtual Image Correlation method for precise silhouette boundary measurement, demonstrating its accuracy, robustness to curvature and contrast variations, and providing tools for uncertainty estimation.

Contribution

It presents a new formulation of the Virtual Image Correlation method that is exact in 1D, insensitive to curvature and contrast, and includes an uncertainty estimation framework.

Findings

Method is exact in 1D and insensitive to curvature.

Bias from luminance variation can be corrected.

Analytical formulas predict measurement uncertainty.

Abstract

The Virtual Image Correlation method applies for the measurement of silhouettes boundaries with sub-pixel precision. It consists in a correlation between the image of interest and a virtual image based on a parametrized curve. Thanks to a new formulation, it is shown that the method is exact in 1D, insensitive to local curvature and to contrast variation, and that the bias induced by luminance variation can be easily corrected. Optimal value of the virtual image width, the sole parameter of the method, and optimal numerical settings are established. An estimator is proposed to assess the relevance of the user-chosen curve to describe the contour with a sub-pixel precision. Analytical formulas are given for the measurement uncertainty in both cases of noiseless and noisy images and their prediction is successfully compared to numerical tests.