Xiaofeng Fan

Abstract

This dissertation presents a novel technique based on Maximization of Mutual Information (MMI) and multi-resolution to design an algorithm for automatic registration of multi-sensor images captured by various airborne cameras. In contrast to conventional methods that extract and employ feature points, MMI-based algorithms utilize the mutual information found between two given images to compute the registration parameters. These, in turn, are then utilized to perform multi-sensor registration for remote sensing images. The results indicate that the proposed algorithms are very effective in registering infrared images taken at three different wavelengths with a high resolution visual image of a given scene. The MMI technique has proven to be very robust with images acquired with the Wild Airborne Sensor Program (WASP) multi-sensor instrument. This dissertation also shows how wavelet based techniques can be used in a multi-resolution analysis framework to significantly increase computational efficiency for images captured at different resolutions. The fundamental result of this thesis is the technique of using features in the images to enhance the robustness, accuracy and speed of MMI registration. This is done by using features to focus MMI on places that are rich in information. The new algorithm smoothly integrates with MMI and avoids any need for feature-matching, and then the applications of such extensions are studied. The first extension is the registration of cartographic maps and image datum, which is very important for map updating and change detection. This is a difficult problem because map features such as roads and buildings may be mis-located and features extracted from images may not correspond to map features. Nonetheless, it is possible to obtain a general global registration of maps and images by applying statistical techniques to map and image features. To solve the map-to-image registration problem this research extends the MMI technique through a focus-of-attention mechanism that forces MMI to utilize correspondences that have a high probability of being information rich. The gradient-based parameter search and exhaustive parameter search methods are also compared. Both qualitative and quantitative analysis are used to assess the registration accuracy. Another difficult application is the fusion of the LIDAR elevation or intensity data with imagery. Such applications are even more challenging when automated registrations algorithms are needed. To improve the registration robustness, a salient area extraction algorithm is developed to overcome the distortion in the airborne and satellite images from different sensors. This extension combines the SIFT and Harris feature detection algorithms with MMI and the Harris corner label map to address difficult multi-modal registration problems through a combination of selection and focus-of-attention mechanisms together with mutual information. This two-step approach overcomes the above problems and provides a good initialization for the final step of the registration process. Experimental results are provided that demonstrate a variety of mapping applications including multi-modal IR imagery, map and image registration and image and LIDAR registration.

Library of Congress Subject Headings

Image processing--Digital techniques; Multisensor data fusion; Remote sensing--Data processing; Optical radar; Signal processing

Publication Date

5-1-2011

Document Type

Dissertation

Student Type

Graduate

Department, Program, or Center

Chester F. Carlson Center for Imaging Science (COS)

Advisor

Rhody, Harvey

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TA1637 .F36 2011

Campus

RIT – Main Campus

Download

Share

COinS