Multilayered neural network models for color blindness

Multilayered neural network models for normal and dichromatic color vision, which realize mapping from cone space into perceived color space by backpropagation learning, are constructed. After learning is completed, the properties of each model acquired by learning, such as spectral response property, hue representation of spectral light, and wavelength discrimination, are examined. Each model predicted the experimental evidence well enough to understand the mechanism. The results suggest that the proposed hypothesis for constructing the dichromatic color vision models is acceptable and also suggest that colors are represented in the visual system by lightness and opposing color responses, that is, redness or greenness and blueness or yellowness