Optimal FFT-based algorithms for referencing multi-spectral images

We live nowadays in an information age, where stockpiles of data are generated every second. Much of this data comes in the form of images, which must be analyzed in order to obtain useful information from them. The problem that we solve in this thesis comes from the field of satellite image processing. When capturing satellite images, we do not know the exact geographical coordinates of the points described by the image. Images need to be georeferenced, that is, referenced with correct geographical coordinates. This can be done by referencing, or aligning, the new image with an overlapping already georeferenced image. In other words, we must find the most appropriate transformation (shift, rotation, and scaling) that aligns one image with the other. There exist algorithms for referencing images; however, these algorithms sometimes do not work well for images captured during different conditions such as different seasons. Satellite images consist of scenes which correspond to different intervals of wavelengths (“bands”). The existing image referencing algorithms only use one of the bands and ignore the information from the others. It is reasonable to decrease the referencing error and solve this “different seasons” problem by using the images corresponding to all available bands. This thesis describes new algorithms for such improved referencing