Partially signal dependent slant transforms for multispectral classification

Abstract

The significant advance in high spatial and spectral resolution sensors technology on board remote sensing spacecraft has increased, by order of magnitude, the spectral bands or features dimensionality of datasets. The need arises, in remote sensing, for feature dimensionality reduction to offset the error in estimation of the class statistics. Principal component analysis (PCA) or KLT the image dependent transform is a useful technique in reducing the dimensionality of the datasets at a high costs of hardware and O(N³) computational operations. This paper discusses the development of partially signal dependent classes of sequential slant Hadamard transforms that require elementary operations and can be used for multispectral classifications.