Estimation of fiber length and diameter distribution from SEM images
Abstract
Image analysis often includes a measurement process. It is well known that direct measurements on images may introduce biases that need to be corrected.
In the case of image analysis of man-made vitreous fibers, one needs to measure their diameter and their length, in order to obtai11 the diameter distribution, which one may want to weight by the length of the fibers, or by their volume. If one is genera.lly able to measure fiber diameters directly after some segmentation steps and correct any measurement bias by Miles-Lantuéjoul-like methods, one cannot access directly the fiber lengths in all cases, for example when both ends of the fibers are not always visible.
In this paper we present three original methods, based 011 different assumptions, that allow to estimate both the unbiased diameter distribution and the mean length by diameter class i11 any configuration, which in turn allow to estiinate with a high degree of confidence any length, surface or volume-weighted diameter distribution. These methods were tested on simulated images, and yielded remarquable results.