Abstract: This paper proposes a new road extraction method of remote sensing imagery based on G statistics of object histogram.Firstly,the original imagery is segmented using marked-watershed algorithm to obtain imagery objects, and the imagery objects' spectral features are extracted,the built-up areas containing roads and buildings are separated based on SVM. Then the appropriate objects are chosen as training samples.G statistics is used to measure the histogram distance between test samples and training samples which describes the heterogeneity of two objects.Minimum distance classifier is employed to separate the road information and buildings in built-up areas. Finally,the final road information is extracted integrating the shape features and mathematics morphology.The result of experiment shows that the proposed method can fairly well extract the road information.

Key words: G statistics, road extraction, watershed segmentation, LBP texture histogram, mathematics morphology