Template for COSC 575 Research Papers

We consider the problem of detecting rooftops in overhead images, which is one processing step in a building detection system. Currently, the system uses a hand-configured linear classifier to select the most promising rooftop candidates for further processing. We present results from an empirical study in which we used machine learning methods to acquire the selection criteria for rooftops. ROC analysis demonstrated that a naive Bayesian classifier performed better than the hand-configured solution, using area under the ROC curve as the measure of performance.