The availability of multiple sensors, big data repositories, and connected users provides a rich set of applications for image fusion. Fusion methods vary and their effectiveness can be compared using quantitative and qualitative approaches. Whereas the previous chapter discussed quantitative methods, this chapter describes qualitative methods.
10.1 Combining Approach, Methods, and Metrics
The image fusion process combines multisource images to provide comprehensive, succinct, and relevant information. There are different fusion approaches and various fusion methods, of which the quantitative metrics described in Chapter 9 can be used to measure the quality of fused images.
The fused images may be efficient for computer analysis, such as fusing visible and thermal images for face recognition. In such a case, the improvement of face recognition, night vision, and biomedical performance should also be effective in increasing human decision making, which is the final assessment of the image fusion usability.
The fused images for human analysis include fused and colored NV imagery (NIR and LWIR) for situation awareness, PET and MRI images for medical diagnosis and treatment, as well as THz and visual imagery for security. The action time and accuracy of identifying potential objects or risks could then be used as an evaluation of the fusion and colorization method.
The fusion evaluation is based on sensors, environments, and targets (e.g., operating conditions, see Chapter 4). It is very challenging to define one universal metric or evaluation method that is suitable for all image fusion methods and applications. However, a general image quality (GIQ) evaluation is still useful for the comparison and selection of fusion algorithms. Certainly, the fusion metrics and evaluation methods must be able to reflect the advantages of a fusion method for the final application.