摘要
本文提出一种新的、简单可行的方法设计自适应的调色板，用具有自组织特性的Ｋｏｈｏｎｅｎ神经网络，根据调色板的要求，设计了具有一维输出神经元的确定结构的网络．同时，为了使主观的显示效果更好，客观的映射误差减小，提出了修改的Ｗｉｎｎｅｒ－Ｔａｋｅ－Ａｌｌ（ＷＴＡ）算法，并对神经网络权值的初始化、输入样本的采样顺序提出了相应的方法．
In this paper, a new, simple and feasible method using self-organized neural network to generate a palette is presented. A definite neural network structure with one-dimension output neurons is developed for this special application. At the same time, a modified learning algorithm of Winne-Take-All(WTA) is proposed for better display quality and less mapping error. The initialization of the weights of the neural network and the sampling method are also discussed in this paper.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
1994年第5期56-62,共7页
Journal of Shanghai Jiaotong University
