Abstract: Landmark extraction is a hotspot of spatial cognition and spatial knowledge organization research. Saliency measurement model is a most important method of landmark extraction, of which the principle is the weighted evaluation of the appearance, structure and semantics of objects. Considering the spatial distribution and geometric characteristics of objects, and it is difficult to extract landmarks with more consistent features, making it a complex process. From the viewpoint of composite features in Voronoi domain, a landmark extraction method is proposed to explore the geometric feature reflection of landmark from individual to its neighborhoods, based on the height of Voronoi element, Voronoi k-order neighborhood crevice and Voronoi k-order neighborhood visibility. Comparative experiments show that: (1) Compared with the landmark extracted by the existing methods with the results of traditional saliency measurement model, which verifies the effectiveness of this method; (2) As for the characteristics of ground property, the geometric features contained in Voronoi diagram can effectively represent the landmarks, reducing the complexity for feature selection of landmarks.