Assessing the spatial accuracy of object-based image classifications