| This dissertation introduces a new framework for image fusion and visualization of nonliteral multiband imagery. Our approach is motivated by the role of contrast perception in the human visual system, and takes advantage of the natural geometric structure inherited by a multiband image through its embedding in photometric space. We first exploit this geometric structure to extend the definition of contrast to multiband imagery, so that we may compare the contrast content of different types of images.; Through the use of generalized contrast, we formulate the image fusion problem as an optimal reconstruction with respect to different energy norms, and provide numerical methods for minimizing those norms. We consider several extensions of the basic method which increase its expressive power and create morphologically correct solutions. Also, we discuss the choice of geometric parameters, both as a function of the imaging sensor and the image noise, in order to optimize the visual quality of the fusion.; After discussing grayscale visualizations, we introduce methods for chromaticity preserving color visualization and hybrid contrast/statistical visualization. Applications in contrast enhancement of noisy images and dynamic-range compression arise as a natural byproduct of the techniques developed, and are discussed in that context. Lastly, we show some partial results in areas which deserve further attention, but in which we have been unable to provide a complete solution, either for lack of a clear optimality criterion or for lack of appropriate numerical methods. |
