Abstract: This paper proposes a novel method of multi-dimensional transfer function design for volume rendering based on RBF neural network. By sampling the characteristic information of interested voxels on the intuitional interactive interface with a brush, this method trains RBF neural network with these training samples. Trained RBF neural network achieves classification and recognition to all the voxels, and different optical parameters are allocated to all the different classification results to indicate, which finishes the design of a transfer function automatically. Experiment results have shown that the interactive interface can define the interested object conveniently and improve the efficiency of human-computer interaction greatly. Besides, the self-learning ability of RBF neural network can avoid the blindness of transfer function designs, enhance the rendering effect of interested regions and achieve the automation and intellectualization of transfer function designs.

Key words: volume rendering, interactive interface, transfer function, RBF neural network

摘要： 提出一种基于RBF神经网络的体绘制多维传递函数设计方法，利用直观的交互界面，通过画笔获得感兴趣体素的特征信息作为训练样本对RBF神经网络进行训练，使用训练后的RBF神经网络实现全部体素的分类识别，对不同的分类结果赋予不同的光学参数进行显示，自动完成传递函数的设计。实验结果表明，所设计的交互界面能直观方便地定义感兴趣的对象，大幅提高人机交互的效率；RBF神经网络的自主学习能力能够避免传递函数设计的盲目性，增强感兴趣区域的绘制效果，实现传递函数设计的自动化和智能化。

关键词: 体绘制, 交互界面, 传递函数, RBF神经网络