Recognition of Reference Microlabeling Images against the Background of Similar Relief Elements

Abstract

Summary. The problem of recognizing a reference microlabeling by microscopic images containing elements of pseudo-labeling (similar to the real substrate relief elements ) is considered. The scope of such marking is the identification of the studied or modified surface areas as well as the lines connecting these areas with macroscopic landmarks on the surface. The microlabeling is formed using a probe microscope cantilever or a nanoindentor. Examples of images with pseudo-labeling elements and the results of their recognition by low-level structural analysis methods previously used for microlabeling recognition are given. In particular, a surface curvature detector was used, which has proven to be good for discrete microlabeling recognition. The effect of pseudo-labeling is the formation of a large number of the image background key points, which reduce the effectiveness of recognition. The application of the linear Hough transform for approximation and subsequent recognition of separate labeling elements is described. It is also shown that to recognize the labelings obtained by the probe microscope cantilever, it is advisable to use morphological erosion before the Hough transformation. The procedure for setting the parameters of this transformation, which most significantly affect the recognition of markings, is described. The range of the recorded Hough transform segments and the Hough transform threshold were used as such parameters. An image processing algorithm and a recognition evaluation criterion are presented. In this case, a histogram of the distribution of the angles of mutual rotation of the segments detected by the Hough transform is used. The recognition criterion is the presence of dominant peaks with certain values in this histogram. The results showing the efficiency of the presented algorithm are presented.