Methods for hyperspectral image visualization and analysis

Hyperspectral data processing and visualization have matured into important research topics. Although new technology in remote sensing and computer hardware has greatly facilitated acquiring and storing the data, visual analysis and georeferenc-ing/registration of the raw hyperspectral data remain hot research areas. One crucial challenge in hyperspectral image visualization is how to achieve multiple goals in the final result. This dissertation contains five contributions to the visualization problem and the registration problem. The first contribution is a new framework which takes two inputs: a user defined target image and a distance metric. The output will be optimized to be similar to the target input as well as preserve the pairwise distances of the input points defined by the user-specified distance metric. The second contribution is a visualization algorithm that casts the problem as a constrained dimension reduction problem. It preserves the pair-wise Euclidean distances between the pixels in the original hyperspectral images as well as maps the original data points to a human-interpretable color space. The third contribution is made in a closely related problem, which is to convert a color image to a grayscale image. A novel mapping algorithm is proposed based on the ISOMAP algorithm. The fourth contribution is a new registration algorithm for hyperspectral image registration. A curve matching algorithm based on a novel similarity transform invariant signature is also proposed for hyperspectral image registration and this is the fifth contribution of the dissertation.