Research On Interactive Multivariable Volume Rendering Based On Subspace Analysis And Multidimensional Projection

Data visualization and visual analysis is a multidisciplinary research area.Its purpose is to display the hidden patterns and valuable information in data.Volume rendering is an important technology for volume data visualization,which has been widely used in biomedicine,combustion and meteorological simulation,geological exploration and other fields.The design of transfer function is a key task of volume rendering technology.It transforms voxel values into optical attributes,which helps users to explore important features in volume data efficiently.The traditional transfer functions are usually designed for single variable,and have single interactive mode.However,the representation of many scientific phenomena often require a variety of attributes,for instance the meteorological data is composed of temperature,pressure,humidity and other properties.The transfer functions of single variable are unable to express the complicated internal structure.Therefore,how to understand and visualize multivariable volume data has become a great challenge.One of the important ways to solve the problem is to understand the relationship between multiple variables,and design the transfer function associated with multiple attributes in space.By adjusting the transfer function,users can interactively analyze and research the complex characteristics of multivariable data.In this dissertation,aiming at the problem of the existing visual analysis methods of multivariate volume data cannot show complex characteristics and are too complicated to operate,starting from the perspective of global and local,proposes a novel multivariate volume data visual analysis method,which combines the subspace clustering method and RadViz(Radial Coordinate Visualization)technique.Due to multivariable volume data generally have large data size,from the global perspective,use K-means++ algorithm to extract representative sample points from raw data.Through subspace clustering method,project the representative high-dimensional sample points into low dimensional space,so as to embody the similarity relationship between different sample points,and at the same time,it is convenient to switch to observe different subspaces.The RadViz technology is improved,the distribution of the corresponding data in different dimensions can be displayed clearly.Meanwhile,the interaction design is carried out to support users to select interesting features and further explore the local details of the selected subspaces.According to the proposed method,construct a subspace analysis and multidimensional projection based interactive visualization system SAMP-Viz(Subspace Analysis and Dynamic Projection Visualization),it can effectively guide users to explore the features of interest and reduce the complexity of operation.The experimental results show that this method can assist users to identify characteristics of complex multivariable volume data effectively and express the distribution information in different dimensions accurately.Our constructed multivariable volume data visualization system SAMP-Viz improves the analysis efficiency and ensures the real-time rendering.