A Brief Introduction to Multivariate Image Analysis (MIA)

Images have been used in the sciences for a long time and they are used increasingly. Large amounts of data representing complex systems can only be represented by visualization as images. Multivariate images arise from a surprising variety of sources. Some are images in the conventional sense (such as satellite data) while others are not (secondary ion mass spectroscopy, SIMS). Almost all physical units can be used to make images and multivariate images: temperature, gravitational field, impedance, magnetic field, electrical field, mass, wavelength, ultrasound wavelength, polarization, electron energy etc. A rough but practical subdivision of the fields of scientific imaging is in satellite imaging, medical (clinical) imaging and the microscopies. The simplest meaningful multivariate image has two pixel indices (e.g. width and height in the image plane) and a variable index, making up a three-way array. An important aspect in going from analog scenes or objects to digital images is resolution. Multivariate images have spatial, intensity, spectral and time (temporal) resolution. A typical older satellite image would have 512x512 pixels, in 7 wavelength bands and an intensity resolution of 256 gray levels. High spatial and intensity resolution is desirable and this makes the arrays rather large and the calculations slow.