Surveying for man-made objects in photographic images is of utmost importance for various military and civilian applications. In this paper, we present two supervised approaches for classifying a photographic image as containing either dominant natural or man-made regions. The first approach has low-complexity where features are extracted from a statistical model of multi-scale sub-band coefficients of natural scenes. The second approach is based on traditional robust feature extraction along with recent deep methods. We evaluate the performance of these approaches on two popular image databases composed of a mixture of man-made and natural scene photographic images. We compare their performance in terms of classification accuracy as well as computational complexity. While the traditional robust feature based classification approach appears to be an obvious choice for such a task, we conclude that low-complexity approaches cannot be discounted for real-time applications. Finally, we also construct a likelihood map for the man-made regions for quick localisation of man-made regions within mixed image that could help in speeding up the detection process. © COPYRIGHT SPIE. Downloading of the abstract is permitted for personal use only.