Appling semantic method to windows detection in façade

摘要

In this paper, the image database and the knowledge representation of the windows in facade are developed to improve the accuracy ratio of Gentle AdaBoost algorithm which provides a final strong classifier in windows detection in façade. To achieve the knowledge representation, this paper had established four windows contextual rules, which consult the concept of image semantics. After the real testing, the results demonstrate the single window detection method is quite successful. It not only can promote the window detection system, but also extend the other relative fields, such as 3D large scales reconstruction and intelligence cars.