Automatic road extraction based on local histogram and support vector data description classifier from very high resolution digital aerial

Abstract

In this paper, an automatic road extraction algorithm for very high resolution aerial imagery based on the local histogram and the support vector data description (SVDD) classifier is proposed. The road maps are extracted by the classification technique. Compared to most existing approaches, the proposed algorithm does not employ the direct stack of spectral bands to form feature spaces, but take the local histogram vector, which calculated from a moving window or a tile of image, as the predictor instead. To evaluate the effectiveness of the algorithm, we apply our algorithm to very high resolution aerial images acquired by Zeiss Imaging's DMC camera. The experimental results indicate that the proposed algorithm is robust and can extract most road pixels correctly even for very complicated scenes.