中文摘要：

在同一个窗口可视化含有多个数据属性值的二维多元数据在很多领域都有重要的应用.为了支持用户在同一个窗口实时交互探索多元数据在不同精度下的可视化结果,实现可伸缩的二维多元数据可视化,提出基于CUDA加速非等轴纹理合成的可伸缩二维多元数据可视化方法.首先通过视觉实验确定纹理样本中纹元的视觉变化与数值变化的对应关系;然后根据实验确定的对应关系,利用非等轴纹理合成方法生成表示二维多元数据变化的纹理可视化结果;再设计了与数值变化对应的用户缩放与平移交互操作,并提出基于CUDA加速的纹理合成以实现用户交互探索;最后针对全球气候数据,给出该方法的可视化结果.用户实验结果表明,文中方法能够有效地完成可视化任务,且优于2个已有方法.

英文摘要：

It is important for many applications to visualize 2D-multivariate data with multiple attributes in one window. In this paper, we present a scalable zoom-independent 2D-multivariate data visualization method. This method not only visualizes the changes of 2D-multivariate data in one window, but also allows users to interactively explore data at different zooming levels. First, we design and perform visual experiments to determine how to map the changes of visual channels of textons to the changes of data attribute values. Next, the example-based anisometric texture synthesis algorithm is adopted to visualize the patterns of 2D-multivate data based on the above visual mapping. Then we design user interactions including zooming and translation to explore the data at different levels of precision. We propose to accelerate the anisometric texture synthesis with CUDA in order to achieve the interactivity of such user operations. Last, our visualization method is applied to the global climate data to get the visualization result of global climate pattern. To verify the effectiveness of our method, we design a user study to compare our method to the other two visualization methods. The results of our user study show that our method accomplishes the visualization tasks more efficiently than the other two existing methods.