Figure 6From: Spatio-chromatic information available from different neural layers via GaussianizationRedundancy reduction (ΔT, in bits/sensor) between the LMS input and the spatio-chromatic representation in V1, estimated via RBIG (top row) and via the theoretical approach of Eq. (9) (middle row). Results are shown for every region of the achromatic contrast/luminance space for different chromatic contrast (in different columns). The surfaces in the blue-yellow colormap are the average of 10 total correlation estimations in each case. The green surfaces at the plots of the top row represent the combination of the uncertainties of the estimates \((\sigma _{\mathrm{RBIG}}^{2} + \sigma _{\mathrm{theor}}^{2})^{1/2}\). The green surfaces at the plots of the middle row represent the absolute difference between the theoretical and the RBIG estimates. The differences are similar to the uncertainty. This agreement stresses the accuracy of the RBIG estimates and the correctness of the theoretical estimates from the analytical model. The bottom row shows the component of the theoretical ΔT that comes from the analytical JacobianBack to article page