Research On Intelligent Design Methods Of Multi-Dimensional Transfer Function In Volume Rendering

Currently, Direct volume rendering has become one of the widely used methods of visualization in scientific computing applications. By processing and outputting all the voxels directly, it can reconstruct a high quality three-dimensional projection image with overall visual effect. As the key step of direct volume rendering technology, the design of transfer function has been always highly concerned. The quality of 3D images is directly depended on optical imaging parameters which are mapped from the voxels of three-dimensional data by the transfer function of volume rendering. However, the problem of volume rendering transfer function design has not been addressed properly, and become the bottleneck of direct volume rendering technology development and application. It is also the core technology and the hotspot in volume rendering researches recently. In this paper, based on the analysis of the design method applied in volume rendering transfer function, by combining RBF neural networks and rendering interface to guide the design process of the transfer function, this research can achieve the automation and intelligence of transfer function design. The research includes:(1). Because of lack of direct interactive interface in current volume rendering transfer function design, an intuitive and rendering based interactive interface is designed. By sampling the characteristic information of interested voxels on the interactive interface with a brush by the users, this method can train RBF neural network with these training samples. Compared with traditional methods, this interactive interface can define interested objects without the priori knowledge of parameter space of transfer function, which will highly increase the efficiency of human-computer interaction.(2). Because of blindness of in current volume rendering transfer function design, based on the deep analysis of the feasibility and effectiveness of current artificial neural network applied on the volume rendering transfer function design, a volume rendering transfer function design method based on RBF neural network is introduced. RBF neural network can be trained by the sampling data constructed from the above interactive interface, achieve the classification and recognition of all the voxels, and different optical parameters are allocated to all the different classification results to indicate, which can finish the design of a transfer function automatically. Experiment results have shown that the ability of self-learning of RBF neural network can avoid the blindness of the transfer function design, enhance the rendering effect of interested regions, and increase the automatic and intelligent level of transfer function design.