Geologists and archaeologist have dealt with the characterisation of marble types for a long time. The identification of this mineral is fundamental to many authenticity testing methods, e.g., it plays an important role in validating the originality of artistic pieces. Archaeologists also use the classification of marbles to identify the provenance of art relics' building stones. In our research, we use a database consisting of images of marbles, including specimens revealed in the ruins of the ancient Troy.
The automatization of the marble recognition has many advantages, such as the reduction of the processing time, since computer based grain identification is much faster than hand-drawing the grain boundaries. Moreover, digitalization improves the accuracy of classification. For instance, histograms of grain size are more informative and distinctive then the average grain diameter used by geologists before.
Our final goal is to develop a program, which helps geologists to classify marbles in a semi-automatic way, partially relying on the user's contribution. In our research we use thin-sections of marble photographed with a microfilm scanner. In the first stage of the classification, the images are processed by edge detector and segmentation algorithms. Inaccuracies of the preprocessing are compensated in the next stage. The user can then improve the retrieved contours using a drawing programme. In the final stage, classification algorithms are applied to identify the type of the marble based on the grain data.
The most crucial task in the process is the separation of the grains, since we use their properties (size, shape) to identify the mineral's type. The aim of this thesis is to present the algorithms we used to identify the marble grains. Several common edge-detecting methods have been applied to localize grain boundaries, however, the results were not satisfying. This deviation is caused by the so-called crystal twinning, which means that two separate crystals might coalesce, resulting in straight lines on the thin-section images. These lines might mislead the common algorithms of segmentation. To overcome these bottlenecks, different image processing techniques are proposed. These methods take into consideration such characteristic properties of the marbles that are ignored by the common schemes.
I have implemented and tested the proposed algorithms in C++, involving the OpenCV class library. The grains reconstructed by the method are in good agreement with the results of human naked-eye examinations. As a perspective, the automatic classification of marbles is targeted.