A General Framework of Image Fusion for Concealed Weapon Detection

Image fusion has demonstrated its effectiveness for remote sensing in general and concealed weapon detection (CWD) in particular. Although a number of research efforts have been made in this area, several issues still remain. First, most of existing schemes perform image fusion at pixel-level and their performances are not always satisfactory. Second, image fusion is commonly considered as a single step process. Third, how image registration accuracy affects the performance of image fusion schemes has not been widely studied. Finally, there still lacks a general image fusion framework. To address these issues, Intelligent Automation, Inc. (IAI) and its consultant, Lehigh University, propose a general multimodal image fusion framework for CWD. The objective of the proposed image fusion framework is to detect concealed weapons as it presents and locate suspect weapon carriers. We present a novel hierarchical architecture, which effectively integrates pixel-level and feature-level image fusion approaches. It achieves high confidence of detection and also avoids the “block” effect of the pixel-level approaches. In addition, the visual information is retained in the composite image for suspect carrier localization. Further, the image registration algorithm is integrated into the framework and a non-reference objective performance metric is proposed.