Natural Texture For Interactive Visualization Of Tweo-Dimensional Multivariate Data

With the fast development of information technology, computer networks and means of data collection, the amount and complexity of data is rapidly increasing which makes it very difficult for us to understand the data of huge volumes. Data visualization technology improves the speed of the people to understand and analyze the data using high bandwidth parallel visual signal processing ability of the human eyes with strong pattern recognition capability. The two-dimensional multivariate data are common in real life, such as weather data, maritime climate data, and oil analysis data. The two-dimensional multivariate data visualization technology can help people to quickly understand the significance of the data and make the right judgments. In this thesis, we study the important research topic of two-dimensional multivariate data visualization based on texture synthesis.We combine the natural texture and the human-computer interaction to present a two-dimensional multivariate data visualization method in this thesis. Using this method, the users can capture the meaning of data by observing visual-friendly natural texture easily and quickly, and interactively displaying data from overall display to the interest details of data. So the users can not only observe the overall data, but also analyze details of the important data. The main work and results of this thesis are as follows:(1) Propose a data visualization method which uses natural textures to encode two-dimensional multivariate data:Natural textures are visually-suitable for human eyes and have high visual dimensions. We present the data visualization method that uses natural textures to encode two-dimensional multivariate data. In this method, the three attributes of two-dimensional multivariate data are mapped to the size, orientation and color of natural textures by adapting the example-based anisometric texture synthesis algorithm. Using this method, the users can understand the data by observing the variance of the synthesized natural textures. We further study how people perceive texture features and identify more appropriate texture variations by designed user study experiments.(2) Propose the acceleration strategies for interactive visualization:Interactive visualization is an important means for users to explore the data and is also an indispensable component of data visualization system. We design and implement the interactive visualization system which enables users to switch between the overall display and the show of interested details interactively. The visualization is accelerated using CUDA, the principal component analysis, the^-coherence search and the super-resolution technology for smooth visualization interactive.(3) Propose a GPU-based fast search of similar patches in images:K-coherence search strategy is needed to preprocess exemplar before accelerating the synthesis. This exemplar preprocess to search similar patches in images is time-consuming. To solve this problem, this thesis proposes a GPU-based fast search of similar patches in images. This thesis uses the parallel search method to accelerate the exemplar preprocess which is faster than the traditional accelerated approaches.(4) The design and implementation of two-dimensional multivariate data interactive visualization system:In this thesis, a two-dimensional multivariate data interactive visualization system was implemented using MFC, CUDA and OpenGL. MFC is responsible for the system’s interface and user interaction; CUDA is responsible for synthesizing the results of visualization; OpenGL is responsible for rendering the results of visualization. The system is divided into four modules:system interface, data processing, visualization mapping and visualization interaction. Each function module is separately designed and implemented, and then integrated into the visualization system. Through this system the user can select the good texture to encode multi-dimensional data, and can control the display of data in different levels of detail to observe the interesting data.