Chromatic structure of natural scenes

We applied independent component analysis (ICA) to hyperspectral images in order to learn an efficient representation of color in natural scenes. In t he spectra of single pixels, the algorithm found basis functions that had b roadband spectra and basis functions that were similar to natural reflectan ce spectra. When applied to small image patches, the algorithm found some b asis functions that were achromatic and others with overall chromatic varia tion along lines in color space, indicating color opponency. The directions of opponency were not strictly orthogonal. Comparison with principal-compo nent analysis on the basis of statistical measures such as average mutual i nformation, kurtosis, and entropy, shows that the ICA transformation result s in much sparser coefficients and gives higher coding efficiency. Our find ings suggest that nonorthogonal opponent encoding of photoreceptor signals leads to higher coding efficiency and that ICA may be used to reveal the un derlying statistical properties of color information in natural scenes. (C) 2001 Optical Society of America.