Image Analysis of Size and Distribution of Particles in Tempered Martensite
Abstract— Image analysis in many fields of science requires a certain level of automation, especially when analyzing a thousands of features on a large amount of images, where manual analysis would be extremely time consuming, if not impossible. The automation level depends mainly on the contrast between the pixels of the features to be analyzed and the background, e.g., carbo-nitrides in a steel matrix. The aim of the present work is to improve the accuracy and shorten the time of automatic image analysis of a large number of precipitates in two grades of steels, X20 and P91. SEM images at different magnifications (3k to 20k) were acquired, depending on the optimal accuracy both in the size and shape of particles analyzed (pixels per particle), and their number density (particles per surface area analyzed). On SE images, precipitates usually appear brighter on a darker background, however a non-uniform gray value intensity in many areas of micrograph is equal for both precipitates and background, which makes it impossible to perform a simple thresholding segmentation. Thus advanced segmentation techniques employing sophisticated algorithms along with image preprocessing in order to improve the segmentation performance itself were applied. It was found that there is no general image processing and segmentation technique, no matter how sophisticated it is, which can cover a wide spectra of different images, even if they might have been acquired from the same material. Thus different combinations of image processing and segmentation algorithms should be applied for each image in particular, in order to achieve acceptable levels of characterization accuracy.