Application of the wrapper framework for image object detection

Tools for automatic image understanding for managing operator workloads are essential. One common task for image analysts is the scanning large collections of real-time images looking for particular objects of interest. This task is difficult to automate due to variable imaging geometries and environmental conditions. This variability of conditions can make automating image strong segmentation for eventual object classification extremely difficult. This paper proposes a tool which integrates image segmentation and classification to allow the integration of semantically meaningful information into the segmentation process. The wrapper framework has previously been shown to be effective in performing strong segmentation on images containing large complex shapes in a fixed field of view. This research extends the applicability of wrapper to wide area surveillance of images containing possibly multiple objects of interest. The approach is demonstrated on aerial images from the Katrina disaster to be able to detect buildings for possible damage assessment.