Acquirement of Categorical Perceptions of Colors by a Neural Network
Details
We can perceive a little difference between colors. But we often use several color names, when we express the colors. Eleven color names ; red, blue, green, yellow, purple, orange, pink, brown, black, white and gray, have been called basic color terms. In order to express color apperance of a certain area, which color terms to use is not only determined by the spectral reflectance of the area, but also influenced by color-apperance modes that change depending on the surrounds. In this report, for the purpose of estimating how categorical color perception is processed in our vision system, we trained a neural network to perceive the color like human by the back-propagation method. After learning was complete, the network approximately categorized unknown inputs into the basic color terms like human. And the responses of three hidden units was like oppornent color response. It is shown that there is a possibility that a neural network will acquire the mechanisms as a human vision system.
