Abstract

Image registration is one of the basic image processing operations in remote sensing. A hyperspectral image has two spatial dimensions and one spectral dimension. There are many hyperspectral sensors used in remote sensing. Traditional intensity-based registration methods may fail for hyperspectral images because of the different spectral sensitivities for different sensors. In addition, not all spectral bands are required to achieve accurate registration. This thesis develops a modification of the large deformation diffeomorphic metric mappings (LDDMM) algorithm in order to deal with the challenges when applied to hyperspectral images. The transformation generated by our method that deforms one image to match the other is differentiable, isomorphic and invertible. We also propose a mutual information based band selection algorithm to reduce the data redundancy of the hyperspectral images. The approach is applied to two hyperspectral images from OMEGA instrument, with a better matching result than original LDDMM method with respect to mutual information.

Committee Chair

Joseph A. O'Sullivan, Miaomiao Zhang, Jr-Shin Li

Committee Members

Joseph A. O'Sullivan, Miaomiao Zhang, Jr-Shin Li

Comments

Permanent URL: https://doi.org/7936/drmj-w167

Degree

Master of Science (MS)

Author's Department

Electrical & Systems Engineering

Author's School

McKelvey School of Engineering

Document Type

Thesis

Date of Award

Spring 5-16-2019

Language

English (en)

Download

Included in

Engineering Commons

Share

COinS