| Image fusion is a process of combining complementary information from multiple images to generate a single image that is more suitable for the purpose of human visual perception or further image processing tasks. The source images can come from different types of sensors of the same basic type or they may come from different types of sensors. In our research, we focus on fusing images from different sensors. The application we mainly considered is the concealed weapon detection application (CWD) which utilizes multiple sensors to improve the identification of the person having weapons hidden underneath his clothing. CWD appears to be a critical technology for dealing with terrorism which is the most significant law enforcement problem for the next decade. In this dissertation, we focused on investigating three aspects of image fusion which have impact on the CWD application, with the hope of shedding more light on this promising area. First, we tested and compared the performance of a large set of typical image fusion algorithms for the CWD application using some typical quality measures. Based on the experimental results, it is clear that there is a need to develop new image fusion algorithms that are more effective for this specific application. In addition, most of the current image fusion work has been limited to monochrome images although the human vision system is very sensitive to colors. Thus, we proposed a new color fusion algorithm. Using the proposed method the fused image will maintain the high resolution and the natural color of the visual image while incorporating any useful information detected by the nonvisual sensor. The utility of the proposed method was demonstrated in the experimental tests. For multisensor image fusion, blind estimation of the image quality is of great interest and of great difficulty. Several objective image quality measures for image fusion have appeared, but there has been a lack of theoretical analysis on the performance of the quality measures. Thus, in this dissertation, we took a first step towards theoretically analyzing one of the popular quality measures, a mutual information based quality measure which had only been tested experimentally for a limited number of cases. We augmented these experimental tests with analysis. |
