Geographical data is being generated at an ever-increasing rate by organisations involved with remote sensing, surveying and mapping. To capture semantic content of this data manually is expensive and so to address this issue, research carried out by our group, the Intelligent and Graphical Systems group of the Department of Computer Science, involves the capture and automatic structuring of data for various kinds of graphical information system. If this task can be automated, it will increase the availability of data and decrease its cost. The goal of the systems currently in place is the recognition and classification of geographical features solely from vector data sets that have been captured through manual digitisation. Modifying the existing classification systems to deal with raster image data (such as remotely sensed images) is desirable for two reasons: Most captured data is initially in a raster data format and must be converted to vector formats. The raster data can contain useful semantic information (such as colour, texture, etc.) that is lost in the conversion. We propose to extend the application and effectiveness of our existing software by incorporating semantic information derived from raster data into our existing vector models: This work can be decomposed into several elements Geometric rectification – align the two data sets. Object extraction – which is releatively simple due to data sets being aligned. Calculate feature vector – to be used in classification. Classification – classify objects with appropriate feature code.