Building Roof Contours Extraction from Aerial Imagery Based On Snakes and Dynamic Programming

SUMMARY This paper presents a method for building roof contours extraction from high-resolution aerial imagery taken over complex urban scenes. The proposed method is based on the optimization of a snakes energy function that represents building roof contours in the image-space digital reference system by using the dynamic programming technique. Formally, snakes (or active contour models) are parametric curves defined in the image domain. Snakes were proposed in the late 1980s and since then they have been very noticeable as one of the most active and successful research topic in image segmentation, and they have been largely employed in image analysis tasks like boundary tracking, shape modeling, and feature extraction. The snakes' position on an image is defined by its energy, which is given by a summation of internal and external energy terms. The internal energy is related to the intrinsic characteristics of the curve that is calculated from the point coordinates defining the curve on the image. External energy, on the other hand, is calculated from image data based on well-defined properties of the object of interest. External energy allows the snakes curve to move towards the object boundary. So, a snake model is able to incorporate different information concerning the desired objects according to the specific application. For the proposed method in this paper, the external energy for the energy function that represents building roof contours uses magnitude for edge points (i.e. a set of points that defines a building roof contour on the aerial image). The energy function is solved by using the dynamic programming algorithm, which is a quite suitable optimization procedure when not all model variables are interrelated simultaneously. In such case, the original mathematical model can be written as a summation of sub-functions depending only on a few model variables. The dynamic programming algorithm advantage (compared to other procedures) is its reduced number of required operations to get the optimal solution (find a set of optimal variables) for the energy function. Furthermore, the dynamic programming algorithm allows imposing constraints to the functional model variables, in such a way that only constrained variables will be selected. As most of the building roof contours are defined by regular polygons, constraints were applied to the energy function to enforce the snakes curve to develop corners on building roof contour corners. Experimental evaluation was carried out using aerial imagery and the obtained results showed the potentiality of the …