Pixels belonging to the same pigments have similar reflectance spectra.

The system is trained by loading spectra of specific pigments and creating the assignement classes, then the software measures each new spectrum and assign it to the closer  class. The distance of spectra is measured with two different techniques. Each spectrum is a vector in a N-dimensional space and each component is the spectral content of each frequency bin. The Euclidean distance is not a valid technique to measure the closeness of reflectance spectra since the change of the intensity of the illuminant would vary the vector length but not the direction and would be seen as a change in the Euclidean distance with respect to other spectra.

Therefore a technique  is based on the measurement of the angle between the new spectrum and the spectra representing each class,  the measured spectrum is assigned to the class having smaller angle. The technique is called spectral Angle Mapper (SAM).

A second technique is based on the measurement of correlation between spectra. The measured spectrum is assigned to the class with higher correlation. The technique is called Spectral Correlation Mapper (SCM).

We have applied the technique to the hyperspectral imagers of the Egyptian coffin measured in collaboration with CCR of Venaria (Turin-Italy). The pigments used by Egyptian artists are basically six: red, green, blue, yellow, white and black. By loading the spectra of this pigments it is possible to assign new pixels to one of the six classes. In Figure 1 we have reported on the right the image in RGB obtained from the hyperspectral image, on the left the RGB in false colors where the different pixels are assigned to black, yellow, red, green and blue.

In Figure 2 we have reported the RGB in false color from a Macbeth colorchecker with only three classes, red, green and blue. It is possible to see that the squares with indexes 13, 14 and 15 are correctly assigned to the right colors. The black color represents the pixels not assigned to any classes.

In Figure 3 we have reported the RGB in false color from a Macbeth colorchecker with only six classes, red, green, blue, magenta, yellow-green and white. It is possible to see that there is a good  correspondance.  The black color represents the pixels not assigned to any classes.

Fig.1, On the left the RGB in false color where the different pixels are assigned to black, yellow, red, green and blue of the . on the right the image in RGB obtained from the hyperspectral image