Page 176 - Computational Colour Science Using MATLAB

10

Multispectral Imaging

10.1 Introduction

The characterization of a digital colour camera so that the device-dependent
RGB values may be transformed to device-independent coordinates such as XYZ
values effectively converts the camera into an imaging colorimeter, and this has
many practical uses. However, in imaging science there are limitations to this
approach. Many applications require that some illuminant-independent
measure, such as the spectral reflectance values, be determined at each pixel
location in a scene and this may be achieved by using an imaging spectro-
photometer. Although imaging spectrophotometers are becoming commercially
available often they are expensive and there is current interest in exploring to
what extent spectral values may be recovered from a standard three-channel
camera system or from a camera system with relatively few channels. The term
multispectral imaging sometimes is used to define this field of research. This
definition is confusing, however, since even the normal RGB image representa-
tion may be described as being multispectral in some sense. In this chapter,
however, the term multispectral imaging will be used to define techniques and
methods that may be used to recover spectral information from camera systems
with a small number of channels (typically in the range 3–8). We distinguish
multispectral imaging from the term hyperspectral imaging which we use to
describe techniques where spectral values are measured using imaging devices
with a large number of channels (typically in the range 16–40). Clearly there are
situations where the distinction between multispectral and hyperspectral may
become blurred. Indeed, we note that the goal of both multispectral imaging and
hyperspectral imaging is the same, namely to recover a spectral image. The
reader is directed to Hardeberg (2001) for an excellent review of this area. We
begin this chapter with a brief review of some computational approaches to the
problem of colour constancy because many of the methods of multispectral
imaging were, in fact, inspired by a computational analysis of the problem of

Computational Colour Science Using MATLAB. By Stephen Westland and Caterina Ripamonti.
& 2004 John Wiley & Sons, Ltd: ISBN 0 470 84562 7