並列型多層ニューラルネットワークによる色識別モデル

概要

論文の詳細を見る
The aim of this work is to implement the human color discrimination system using neural network simulation. ln consideration of human biological system, our model is based on the following four points ; 1 : There are three kinds of cones and one rod in human retina. 2 : The cones and rod are linked at holizontal cell. 3 : The man percieves the colors categolically. 4 : The adaptation system which adjusts the sensitivity of cones and rod plays the important role in the process of human color perception. We have developped a new type neural network named paralell Multi Layers Neural Network which consists of two inputs and two first hidden layers such as paralelle line. These two lines connect at the second hidden layer. In the simulation. spectral reflectance distribution of color tips was used as input signal, and for teacher signal, the Japan Industrial Standard (JIS) No. Z8102 : Names of Non-Luminous Objects Colors was used. For learnning algorithm of this network, backpropagation method was chosen. After learnning, this network can discriminate more than 10 pairs of spectral reflectance distribution and corresponding name of JIS Z8102.
東京工芸大学の論文
1993-01-31