Vibrational spectroscopic imaging has already demonstrated enormous potential for studying a variety of chemical and biological systems at both the microscopic and macroscopic level. However, these spectral images are large and complicated, typically consisting of tens of thousands of pixels, each with an associated high-resolution vibrational spectrum, leading to data sizes upwards of 64 megabytes. In order to realize the full potential of these spectral images, we must find ways to query the data so that specific questions can be answered.

We illustrate a multivariate approach to this challenge, where each pixel is considered to be a single point in a multivariate (N-dimensional) space. The variables (coordinates) of the point in N dimensions are simply the intensities of the N-point spectrum associated with the pixel. In this representation, pixels with similar spectra will tend to cluster together in the multivariate space, since they will have similar coordinates.