Conventional color imaging has three channels—R, G, and B. In multispectral imaging within the visible spectrum, the number of channels increases in order to improve color accuracy and estimate spectral reflectance factor. Image quality criteria important in multispectral imaging
include colorimetric accuracy, sharpness, registration, and low noise. The color transformation matrix, connecting camera signals with CIE tristimulus values, affects color accuracy and the visibility of image noise and misregistration when the multiple channels are combined to a colormanaged
image. When the final goal is a color-accurate image for one set of illuminating and viewing conditions, the color transformation is often derived directly using nonlinear optimization minimizing the average color difference between spectrophotometer- and camera-based colorimetric coordinates.
Optimization requires starting values and least squares minimizing spectral or tristimulus RMS error is typically used. Although it is effective for achieving convergence, the optimized matrix can result in a large reduction in image quality caused by noise propagation via the color transformation
matrix. These concepts are reviewed.
Investigations on infrared-channel-image quality improvements for multispectral imaging
