Supporting semi-automatic marble thin-section image segmentation with machine learning

For archaeologists knowing the provenance of mar-ble artifacts is important. The methodologies are based on finding the boundaries of the marble grains but only a few algorithms are available to do this instead of the expert. In this paper we propose an adaptive algorithm, called live-polyline, which is able to help the experts marking the grain boundaries and it is able to learn from user interactions as well. We investigate two different approaches. The first one is a heuristic based method, however the other one is a machine learning based solution. We define metrics for the performance, identify its key indicators, provide an algorithm to calculate it and determine the required values of the key indicators for sufficient performance. We also examined the heuristic and machine learning methods in terms of these indicators and measured their performance.