A group of geologist and geophysicist is busy with categorising marble based on its certain properties. The reason behind this can vary. It is imaginable, these attributes reveal which marble piece comes from which quarry or based on this information whether the group deals with forgery.
Firstly, they cut a so called thin section from the given marble piece in the thickness that a scanner can shrine through it. After the scan from the resulting picture they try to specify the properties of the thin section.
The trouble is, the picture revealed before them contains markings of many hundreds or thousands of calcite grain and they have to process all of them to gather reliable data. Unfortunately, current fully automatised image analysers don't work reliably because of the unique structural properties of marble; manual processing, in turn, is too slow.
Therefore a thought is born, a program should be offered to geoscientists which with the help of limited manual interaction can help its underlying algorithms to carry out feature extraction.
In this paper we introduce the GrainAutLine framework which encompasses these algorithms and provides a uniform programming interface for external developers while granting a well-organised technical documentation. Moreover, the framework provides an easy to use user interface for geoscientist users who are able to manually correct the inaccuracies of the algorithms.
The presented system is kind of a special drawing program, and as such, its fundamental part is image visualisation. An average thin section image is typically big in size, above 20 mega pixels, thus fast processing of pixel data is required. In order to achieve this, a video hardware accelerated, NVIDIA CUDA based approach is introduced.