An Adaptive Active Contour Model for Building Extraction from Aerial Images

Abstract

Building extraction from aerial images is one of the recent topics of remote sensing used in many applications such as urban planning, disaster management, military planning, and Geographic Information Systems (GIS). One of the commonly used approaches in building extraction is Active Contour Model (ACM), also called snake model, for its ability to extract contours of structured and unstructured shapes of objects. However, using traditional ACM model in building extraction faces the problem of narrowly concave contour regions. In this research, we propose to solve the deep concavities problem with the use of a concavity index which adaptively determines the rigidity coefficient of the snake points located in the deeply narrow segments of the contour. Our adaptive model was tested on different sets of buildings extracted from aerial images. Results were evaluated using two evaluation approaches. One in terms of accuracy, precision and recall, and the other in terms of the Error Distance Ratio (ERd) which is the average ratio of distance between each snake point and the true edge map point (by pixels). Result were compared with the GVF snake model in terms of both accuracy and execution time.