Spectral imaging uses a computational approach to visually separate overlapping fluorescent spectra.  This is accomplished by one of two methods, "linear unmixing" or "spectral fingerprinting".  In both methods, a "lamda series" is collected by scaning the sample with a laser in standard fashion, but the emission light from every pixel is spread into its component wavelengths with an optical grating or prism.  With the Zeiss Meta system, every 10 nm of wavelength is sent to a separate PMT.  A maximum of 32 channels, over a range of 320 nm can be collected.  Up to eight channels can be displayed. 

In the illustration to the left, each space between the red lines represents an image collected at a specific 10 nm-wide wavelength band.  Four of the five "lamda" images would result from the contribution of two separate fluorescent dye spectra, while the last would be from only one dye. 

Image intensity is a linear combination (mixing) of all contributing fluorescent components.  Linear unmixing uses algebraic algorithms to separate the contribution of each component.   The algorithm uses weighting based on spectral characteristics of each dye involved.  Those characteristics are extracted from regions of the sample known to exhibit just one of the dye spectra. This routine enables the system to identify how much each source contributes to the overall signal at each pixel.

With spectral fingerprinting, pre-existing spectra of the dyes in question are called from memory and used to identifythe spectra in the lamda scans.  This can be done while scanning the sample so results are immediate.  The main differences between the two methods are that linear unmixing uses information contained in the captured image and displayes the results after image acquisition, .  Spectral fingerprinting starts with known spectral profiles and processes the image as it is captured.  They each have their limitations and should be used with caution with fluorphores whose spectra overlap broadly.