Synergistic Fusion of Hyperspectral and High Resolution Image for Improving Performance and Reliability of Automatically Extracted Urban Features

Abstract

Image fusion is a generic word referring to several techniques of digital image processing which are used to
integrate data from different spatial and spectral resolutions in order to obtain higher-quality synthetic images.
This paper emphasizes the assessment and systematic analysis of image fusion techniques by measuring the
quantity of enhanced information in fused images. EO1- Hyperion and IKONOS (MSS+PAN) have been fused
using Principal component analysis, Gram-Schmidt Transformation (GST) and High Pass Filtering (HPF)
algorithms. The photo interpretive potential and their statistical ability to preserve the spectral quality of fused
data, in comparison with original Hyper-spectral image, have been investigated. A set of measures of
effectiveness such as Correlation Coefficient, Mean, Median, Standard Deviation, and RMSE are used for
comparative performance analysis and then best of fusion algorithms has been used for the purpose of
automatic extraction of various urban features. This paper also explores the utility of object-oriented method to
extract various urban features from the best fused high resolution data. The results were evaluated through
comparison to manually acquired data. Several quality measures (Completeness, Correctness, and Quality etc.)
were used for evaluating the accuracy of extraction. The results indicate that the fusion of Hyperspectral data
with high spatial resolution data has an edge over multispectral dataset in terms of automatic extraction based
on roof and road material.
Keywords: Hyperspectral, resolution, urban features, RMSE, sensors