Image analysis of Light Optical Microscopy (LOM) images of metallurgical samples

Abstract: Processing metals in different ways often requires stringent quality control to ensure the quality of the processed metal. This master's thesis aims to develop methods for evaluating the quality of metals via image analysis rather than inspecting images manually as it traditionally has been done. In this thesis methods are developed to solve two problems, both relating to measuring the amount of non-desirable content in Light Optical Microscopy (LOM) images. In the first problem all the non-metallic inclusions in a steel LOM image is found and the width, height and area of each inclusion is stored. The focal point of this problem is removing the appendages of the inclusions and comparing different methods for doing so. In the second problem images containing cracks are inputed and information regarding the cracks present in the image is calculated and stored. The information stored is the number of cracks, the combinde length of all cracks, and the length of the longest crack. The focus in the second problem is in developingmethods to extract the cracks while removing the background features. The methods developed to solve the two problems were either based on traditional image analysis methods (morphological operations for example) or methods specifically designed for the problem (taking advantage of specific patterns that are present in the problem). The different methods were comapared and typically methods designed for the problem worked better than generall methods adapted to the problem. The results of the methods developed were either evaluted bycomparing images, comparing the results to reference data (methods were adapted to work on Spectrum Electron Microscopy images in this case) or by asking an expert.