Semantic segmentation of settlement patterns in gray-scale map images using RF and CRF within an HPC environment

The segmentation of complex and compound geographical objects such as settlement areas in images is a challenging task. This is particularly true if the images are scanned historical maps, which are only available as gray-scale images. On the other hand, their vast availability and spatio-temporal coverage make these maps a valuable source of information for historical land use and land cover (LULC) research and urban geography. To facilitate the laborious information extraction from these maps, we present a two-stage machine learning-based approach for segmenting settlement and non-settlement areas in the map scans. We employ a Conditional Random Field (CRF) which obtains its unary potentials from a Random Forest (RF). The method is tested using two inference algorithms. To evaluate the performance and the scalability of the approach over large amounts of data sets, we conduct parallel computing experiments within a High Performance Computing (HPC) environment at the Center for Information Services and High Performance Computing at TU Dresden. Experimental results indicate the suitability of both the methodological approach as well as its parallel implementation.