Finding Regions of Interest for the Detection of Man-Made Objects in Non-Urban Scenes

This report presents a new approach for the detection of large man-made objects in a rural area using a single monochrome image. In this problem, man-made objects may be unspecified and the appearance of the objects is unpredictable. Prominent features discriminating man-made objects from natural objects are identified. A computational framework for applying perceptual organization and using the prominent features is presented. Techniques are developed to group low level image features hierarchically into a region of interest (ROI) likely to contain man-made objects. These techniques include linear structure extraction, primitive structure formation, and region of interest location. Each of these methods presents its own unique property and advantage as compared with previous related work. Experimental results are presented using real images that have different kinds of man-made objects and a complex background. We show that the located ROIs properly enclose the man-made objects in the scenes. The search space is, therefore, reduced from the whole image to the ROI. •This research was supported in part by the Army Research Office (contract DAAL03-91-G-0050), and in part by the General Dynamics (purchase order WPC001023).