GrainAutLine marble thinsection image processing
The new GrainAutLine website is up and running: http://bmeaut.github.io/grainautline/index.html.
GrainAutLine is an application developed for the semi-automatic image processing of marble thinsections. The primary application is to extract the grain boundaries in an "as automatic as possible" way. The grain outlines are used for classification purposes in archeometric applications. The first task the system is being developed for is the identification of the origin of marble samples found in the ruins of the ancient Troi.
As finding the grain boundaries is a hard task due to twin crystals, GrainAutLine aims for a semi-automatic approach: it is a special drawing program providing as many functions for grain border detection as possible. As the system is developed, it can help more and more the user, so that the amount of manual corrections decreases continuously. Such functions are for example the automatic cut of grain candidates at narrow border discontinuities, or the automatic merge of grain candidates probably separated by twin crystal lines. This way, even during the manual corrections, the user only needs to guide the system and not to draw the detailed borders.
The solution consists of the following main parts:
- Image segmentation aiming to identify the grains and avoid the detection of lines caused by twin crystals.
- User interface providing highly ergonomic functions for the experts, so that the manual corrections can be performed in a convenient way.
- After the borders are identified, machine learning techniques are employed for the marble classification. This step is based on several properties extracted using the grain borders and several other parameters extracted from the images. Some examples are grain size histogram, color, and convexitiy of the grains.
The evaluation of the system is performed with the unique MissMarble database (http://missmarble.de/)
For further information, contact Kristóf Csorba.