Along with the progress of the content-based image retrieval research and the development of the MPEG-7 feature descriptors, there has been an increasing research interest on object recognition and semantics extraction from images and videos. In this paper, we revisit an old problem of indoor versus outdoor scene classification. By introducing a precision-boosted combination scheme of multiple classifiers trained on several global and regional feature descriptors, our experiment has led to better results compared with previous approaches.