ABSTRACT
This paper presents a new analytical method to determine interface normals from a series of bright/dark field images taken from arbitrary orientations. This approach, based on a general geometrical model of interface projection, provides a generalized formulation of existing methods. It can treat an excessive number of inputs, i.e. orientation conditions. Given 6 or more sets of inputs, even with considerable experimental errors, we prove that this method is still very likely to yield results with satisfactory accuracy. The robustness of the method can thus allow its implementation in problems dealing with a large amount of data. We show that this method can also be applied to determine 1D features or to check the planarity of microstructural features.
ABSTRACT
Boosting is a family of supervised learning algorithm that convert a set of weak learners into a single strong one. It is popular in the field of object tracking, where its main purpose is to extract the position, motion, and trajectory from various features of interest within a sequence of video frames. A scientific application explored in this study is to combine the boosting tracker and the Hough transformation, followed by principal component analysis, to extract the location and trace of grain boundaries within atom probe data. Before the implementation of this method, these information could only be extracted manually, which is time-consuming and error-prone. The effectiveness of this method is demonstrated on an experimental dataset obtained from a pure aluminum bi-crystal and validated on simulated data. The information gained from this method can be combined with crystallographic information directly contained within the data, to fully define the grain boundary character to its 5 degrees of freedom at near-atomic resolution in three dimensions. It also enables local atomic compositional and geometric information, i.e. curvature, to be extracted directly at the interface.
