Realization of Color Sensibility Using Neural Networks

近年来，“人的感性”在设计、人机界面、计算机图形学、计算机音乐、演奏机器等各个领域的作用越来越重要。影响人的感性的因素有很多。特别是颜色，强烈地影响着我们的思想，唤起各种各样的形象。本工作的目的是实现一个人类对颜色的感知系统，从电视摄像机获得的R（红）、G（绿）、B（蓝）值中提取“喜-恶”、“暖-凉”等图像作为颜色信息。该系统采用神经元任意连接的神经网络，由两部分组成。其中一种是基于色觉的生理和心理知识构建的色觉子网络。另一个是感性本身的另一个子网络。神经网络从感官测试获得的教学数据中学习人类的色彩感知功能。学习后的网络输出类似于我们脑海中的彩色图像。通过学习后对隐藏单元的分析，证明了以下结果。色觉子网络的反应与传统色觉模型相似。敏感子网络的隐藏单元选择性地响应“Hue”和“Tone”。输出单元整合了这些选择性响应，因此网络实现了对颜色的敏感性。

Recently, "Human Sensibility" is becoming increasingly important in various fields, such as design, human interface, computer graphics, computer music, and playing machine. There are a lot of factors which influence human sensibility. Color, in particular, strongly works on our mind and rouses various images.The purpose of this work is to realize a human sensibility system for colors which extracts some images such as "pleasant-unpleasant" and "warm-cool" from R (red) , G (green) , and B (blue) values obtained with a TV camera as color information. The system is constructed using a neural network in which neurons are arbitrary connected and consists of two parts. One of them is a sub-network for color vision which is constructed based on its physiological and psychological knowledge. The other is another sub-network for sensibility itself. The neural network learns a human function for color sensibility from teaching data obtained by a sensory test. The outputs of the network after learning are similar to color images that we have in our mind.The following results are proved by the analysis of the hidden units after learning.1.The sub-network for color vision shows similar responses to those of a conventional color vision model.2.The hidden units of the sub-network for sensibility responds selectively to "Hue" and "Tone". The output units consolidate those selective responses and consequently the network realizes the sensibility for colors.