Multisensor Image Registration Utilizing The LoG Filter and FWT

摘要

This thesis examines the utility of automated image registration techniques developed by the author. The major thrusts of this research include using the Laplacian of Gaussian (LoG) filter to automatically determine ground control points (GCPs) and wavelet theory for multiresolution analysis. Additionally, advances in both composite and predictive transformations will be covered. The defense will include an overview of the processes involved in general image registration and specifically how they pertain to automation with the techniques utilized in this thesis. Use of the LoG filter to extract semi- invariant GCPs, development of automated point matching schemas, and the use of matrix transformations for efficient management of affine image relationships will be explained in detail. Additionally, the ability to apply statistical analysis to both local and image wide sets of GCPs will be discussed. The student developed software application, LoG Wavelet Registration (LoGWaR). will demonstrate the utility of these techniques for processing large datasets such as LANDSAT and how integration of these features can provide both power and flexibility when registering multiresolution and/or multisensor images. Automation techniques will be highlighted, demonstrating the strengths and weaknesses when applied to images with high degrees of parallax, cloud-cover, and other types of temporal change. Specific applications, such as wavelet sharpening and spectral unmixing will be addressed as it pertains to current research.