Random sampling-based relative radiometric normalisation
Date
2021
Authors
Bonnet, Wessel
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Relative radiometric normalisation (RRN) is a widely used methodology for change detection and radiometric calibration of corresponding multispectral images for further analysis. However, standard RRN methods are not robust against anomalous (or outlier) pixels, which warp the calibration and decrease the spectral similarity of processed images. This research seeks to improve the calibration of corresponding multispectral images through relative radiometric normalisation by utilising a novel random sampling-based method based on the random sampling consensus (RANSAC) to exclude outlier pixels from the analysis. A comparison is made against the widely used Covariance Equalisation (CE), Multivariate Alteration Detection (MAD), Iteratively Reweighted MAD (IR-MAD) and Iterative Slow Feature Analysis (ISFA) algorithms in terms of computing times, mean squared error and the structural similarity index measure. The experimental results show that the proposed method performs favourably against CE, MAD, IR-MAD and ISFA in all metrics considered in this research
Description
A research report submitted in partial fulfillment of the requirements for the degree of Master of Science in the field of e-Science to the School of Computer Science and Applied Mathematics, University of the Witwatersrand, 2021