Interactive and Automated Segmentation and Generalisation of Raster Data

This paper review past and current research activity in the area of generalisation of spatial data and presents a new methodological framework for segmentation and generalisation of raster data. In order to overcome drawbacks associated with supervised classification and generalisation of raster data, an Interactive Automated Segmentation and Raster Generalisation Framework (IASRGF) was developed and tested. Test results of the IASGRF shows that all objects derived from the generalisation of landuse data over Canberra, Australia, were well classified and mapped. The error assessment indicates that the percentile classification accuracy is 85.5%, whereas the commission error is relatively high (38.5%). More importantly, the maximum likelihood classifier using training sites and associated ground truth data suggests that the Kappa index is 0.798, which can be interpreted as a reliable and satisfactory classification result. In order to further enhance supervised classification, a post-classification was carried out. As a result, this extra process improved the overall classification accuracy slightly, however its commission error also increased by 6%.