Based on the intensives study of semantic similarity algorithms and artificial neural networks knowledge, a generalized regression neural network semantic similarity algorithm is proposed. Training samples are obtained by extracting the principal component of semantic similarity influence factors; the desired spread factor and best training sample sets are gotten by cross-validation and recursive optimization; a generalized regression neural network is established with these supports. Experiment comparison and analysis verify that, the result of semantic similarity algorithm based on generalized regression neural network is more accurate than that of existing algorithms.

While it is very reasonable to use a multigraph consisting of multiple edges between vertices to represent various relationships, the multigraph has not drawn much attention in research. To visualize such a multigraph, a clear layout representing a global structure is of great importance, and interactive visual analysis which allows the multiple edges to be adjusted in appropriate ways for detailed presentation is also essential. A novel interactive two-phase approach to visualizing and exploring multigraph is proposed. The approach consists of two phases: the first phase improves the previous popular works on force-directed methods to produce a brief drawing for the aggregation graph of the input multigraph, while the second phase proposes two interactive strategies, the magnifier model and the thematic-oriented subgraph model. The former highlights the internal details of an aggregation edge which is selected interactively by user, and draws the details in a magnifying view by cubic Bezier curves; the latter highlights only the thematic subgraph consisting of the selected multiple edges that the user concerns. The efficiency of the proposed approach is demonstrated with a real-world multigraph dataset and how it is used effectively is discussed for various potential applications.

Aiming at simplifying and real-time rendering of large-scale terrain, the whole terrain is first divided into terrain blocks, and for each block, feature points are picked out to generate a Triangulated Irregular Network (TIN) model. Then, the hierarchical multi-resolution model of the terrain is created in a bottom-to-up mode. Cracks between terrain blocks can be easily remedied by boundary sharing. At run-time, whether a terrain block is to be rendered or not depends both on its isotropic error metric and the result of view frustum culling. Experimental results show that the proposed algorithm performs efficiently.

Computer generated clouds are indispensable in virtual environments nowadays. This paper proposes a probability fields controlling cellular automata method. Creation of the probability fields adoptes fractional Brownian motion (fBm) which accords with the natural chaotic motion of particles and endows the results with realistic cloudy characteristics. We accelerate our simulation by running all the modeling procedures on the programmable graphics hardware. The illumination model accounts for the sunlight and background lights with single scattering model. We use ray casting technique to render 3D realistic clouds. Integration of our results into virtual terrain environment enhances the reality of the scene. The total virtual environment is rendered at interactive frame rates.