Image Segmentation for Reflected-Light Microscopy: Some Theoretical Approaches

Often when we approach the study of lithologies coming from the urban environment, but in general, even from any other environment, be it a quarry, a mine, an outcrop of our interest, the first study we carry out is the one in reflected-light optical-microscopy. Reflected-light microscopy in respect to transmitted-light microscopy requires minimal sample preparation, having to polish a single surface and without the need to thin the samples to allow light to pass through them. It may be useful, already in the first analysis, to try to produce quantitative data on what we are observing. A further advantage of reflected light in an urban environment is that of being able to observe and describe the formation or interaction between opaque minerals and the environment. Information that we lose by passing directly to the transmitted light. The information that can be useful to us and that we can obtain are the relative porosity of the sample, the texture (when easily recognizable in reflected light), the maximum size and shape of the edges of the grains. To all this is added the relationship between the areas of the different crystallites identified and the possible background mass, which cannot be solved on the observation scale. When we are dealing with many samples, we do not always have the time to be able to study individually sample by sample through images, so we resort to the use of image analysis tools for image segmentation and analysis. Among these, the main thresholding method with the Otsu method, the segmentation with the k averages algorithm, and, finally, a neural network of the SOM type. In this short work, we will review the main methods of image segmentation plus an innovative method developed by our group, highlighting its strengths and weaknesses.