Dependence of texture classification accuracy on spectral information

The influence of the number of spectral channels of spatial information on the accuracy of texture classification was evaluated by modelling the spectral sensitivity functions of a group of ideal imaging systems with a set of gaussian functions and then applying them to a set of hyperspectral images in order to simulate the response of each of their colour channels. Feature extraction and classification with different techniques were applied to the simulated data to assess the performance. It was shown that a significant enhancement of the accuracy was achieved, with a dependency on the approach adopted; it is therefore possible to estimate the optimal number of spectral channels for each method.