Clustering Algorithms Applied to Spectral Reflectance Recovery from Simulated RGB Camera Responses

Clustering techniques have been applied to spectral images mostly to classify materials or objects included in the scenes. We present in this study new data concerning the use of clustering algorithms as a preprocessing step for obtaining spectral data from camera responses. Four different clustering algorithms have been used, and the Kohonen maps seem to provide very promising results for spectral recovery based on simulated digital counts wihout or with a high level of additive noise included. We have considered also the how the addition of a wide-band color filter in front of the camera lens to increase the number of channels of the camera from three to six affects the results. The influence of the number of classes for all four algorithms and map configurations for the Kohonen map have been studied as well.