Multispectral Imaging

The VASARI system uses multispectral imaging in image acquisition. The multispectral imaging involves using carefully selected optical filters to capture the whole spectral information instead of the conventional three colorimetric values. It is well known that it is possible to obtain a fairly good reconstruction of the color tristimulus values of the reference human observer, as defined in colorimetry, with 3 well-chosen filters. Multispectral imaging aims to reconstruct the spectral reflectance curve using more than three filters.

The components of the acquisition system include the spectral radiance of the illuminant l, the spectral reflectance of the object surface imaged r, the spectral transmittance of the optical filter and optical path o, and the spectral sensitivity of the sensor a. These determine the camera response of the system, which is

Estimating spectral reflectance from camera responses is done by applying approximation algorithms to the illuminant independent sample values. Some reconstruction algorithms that are used are splines, Modified Discrete Sine Transform (MDST), Modified Discrete Sine Transform with aperture correction (MDSTA), pseudo inverse, smoothing inverse and Wiener inverse. These methods are suitable for filters having rather narrow bandpass shape and being located at approximately equal wavelength intervals as used in existing multispectral acquisition systems. They are not well adapted to filters having complex wide band responses and suffer from aliasing errors.

The choice of filters is important to the quality of the spectral reflectance reconstruction. Some authors have proposed designing optimal filters given an optimization criterion, others have proposed using a set of heuristically chosen filters and yet others have proposed using filters chosen from a set of readily available filters with optimization.

After multispectral data is obtained by scanning, it has to be transported to the output device. An optimal encoding format needs to be compact so as to save transportation costs and needs to be compatible with the three-channel technology of today. A simple way to achieve this is to transmit the image twice, once as a conventional three-channel image and once as a multispectral image. This way is obviously not compact. A better way to encode the information is to have both the three conventional color values and additional multispectral data and hide the additional information from three channel devices for compatibility.

Applications that use multispectral imaging include digital imaging projects such as VASARI.