Abstract:
The application of Mathematical Morphology (MM) techniques
has proven to be beneficial in the extraction of shapebased
and texture-based features during remote sensing image
analysis. The characteristics of these techniques, such as nonlinear
adaptability and comprehensive lattice structure, make
them useful for contextual spatial feature analysis. Despite
the advancements, there are still persistent challenges, including
the curse of dimensionality, maintaining spatial correlation,
and the adaptability of morphological operators in higher dimensions.
The focus of this thesis is to explore the potential
of MM-based methods to analyse spatial features in addressing
these challenges, specifically in the context of spatialcontextual
feature analysis of hyperspectral images and Digital
Elevation Models. This thesis explores the power of morphological
distance in capturing spatial relationships and proposes
a modified definition called "Dilation Distance" to address the
"Dimensionality Curse" in hyperspectral images. By employing
dilation-based distances, spatially separated objects can be
identified, reducing redundancy and enhancing efficiency. Experimental
trials demonstrate the superiority of the proposed
approach. Additionally, the thesis introduces a new approach using morphological interpolation for terrain surface interpolation,
preserving geometric structure while providing a smooth
surface. The extension of conventional univariate morphological
tools to hyperspectral images in a multivariate way is also
explored, ensuring the concurrent application of operators while
preserving the multivariate nature of the data. To achieve that
a vector ordering strategy is proposed. Overall, these techniques
have a profound impact on the progress of mathematical
morphology in remotely sensed image analysis, offering valuable
insights.
