Learning vector quantization for road extraction from digital imagery

Many operations require the most accurate and complete topographic information available. Typically map products cannot maintain currency because of the rapid pace of development. Hence, there is an urgent requirement to exploit satellite imagery to provide current topographic feature data. Among the most important features needed are roads and, hence we require automated procedures to rapidly identify road networks in imagery. This paper describes the use of learning vector quantization to extract roads from digital imagery. We provide results using data from SPOT imagery.