Multivariate and Hyperspectral Image Analysis

Multivariate image analysis (MIA) and hyperspectral image analysis (HIA) are methodologies for analyzing multivariate images, where the image coordinates are position (two or three dimensions) and variable number. Multivariate/hyperspectral images can have typical sizes 1024 × 1024, 512 × 512, 256 × 256, etc. and have between two and many hundreds of variables. The variables can be wavelength, electron energy, particle mass, and many others. Classical image analysis concentrates mainly on spatial relationships between pixels in a gray level image. MIA/HIA concentrates on the correlation of structure between the variables to provide extra information useful for exploring images and classifying regions in them. The many variables can be transformed into a few latent variable images containing condensed information. The sheer size of the data arrays necessitates visualization of raw data, intermediate data, model parameters, and residuals. When the images consist of continuous spectra having more than 100 variables, the name hyperspectral is used and the emphasis of the analysis is often on the spectral interpretation. All physical techniques for measuring materials can be made into imaging techniques, describing not only a property but also its position in a plane or volume. All imaging techniques can be expanded to become multivariate/hyperspectral. Multivariate imaging is used in many fields of research, but for practical reasons a subdivision in the classes remote sensing, medical imaging, and microscopy (including macroscopy) can be made. In microscopy, MIA/HIA can be used to study materials and biological processes by optical, electron, and charged particle techniques.