Detection of streets based on KLT using IKONOS multispectral images

In this paper we propose a target detection approach, in order to detect streets, using an IKONOS multispectral image, with 3 spectral bands, based on KLT - Karhunen-Loeve transform. Our approach is trained to track for asphalt areas, with the aim of finding out the asphalt streets on the image. The approach performs dimensionality reduction, using only the eigenvectors with the highest eigenvalues, generating an eigenspace of low dimension. The target detection is done finding out the shortest Euclidean distance among the primitives of the new images and the primitives of the class we are working with.