Using an ensemble of spectral images that represents a good cross-section of naturally occurring images, Ruderman et al. proceed to decorrelate these axes. Their motivation was to better understand the human visual system, which they assumed would attempt to process input signals similarly. We can compute maximal decorrelation between the three axes using principal components analysis (PCA), which effectively rotates them. The three resulting orthogonal principal axes have simple forms and are close to having integer coefficients. Moving to those nearby integer coefficients, Ruderman et al. suggest the following transform: