Research And Implementation Of Parallel Strategy Of Feature-based Streamline Seeding

Streamline Visualization is an important method for visualizing vector field visualization with characteristics of small amount of computation, intuitiveness and interactive analysis, and is widely used in engineering practices. As the visual effect of streamline visualization depends on the number and location of seedpoints, seedpoint placement has been a hotspot of the research of streamline visualization. To better understanding the feature of the flow field, we usually use the streamline seedpoint distribution of feature-based methods, but contemporary approaches are too slow to meet the needs of real-time interactive visualization. This paper studies two parallel acceleration algorithms which seed streamlines based on features, and preliminarily designs and implements a software with feature-based streamline parallel visualization. The main work of this paper can be summarized as follows:1. We present a parallel acceleration method of similarity-based streamline seeding. We first generate an alternative set of seed points, then each of the threads obtains seeds and integrates to grow streamlines in parallel. With a similarity-distance constraint, every streamline generated by threads affects each other. A copy-and-cache technique is proposed to avoid read-write conflicts and pauses amidst threads, and finally achieves a well-placed streamline distribution constrained by similarity-distance. The experiments show that this method can take advantage of parallel computing performance of stand-alone multi-core, acquire higher parallel speedup, and effectively speedup the generation of streamlines.2. We present a parallel acceleration method of seedpoint placement based on information entropy. We first obtain the characteristic area of the vector field by information entropy theory, and get the initial set of streamline seedpoints. Then we use the copy-and-cache technology to grow streamlines in parallel; after fetching the initial streamlines, we implement parallel approach of vector field block division, using generated streamlines to reconstruct vector field, and then exploit conditional entropy to get more seed points. In the end, we get the streamline visualization that highlights the key feature of the vector field. Experimental results show that this method can obtain higher speedup, and improve the efficiency of the streamline visualization significantly.3. On the basis of the work above, a feature-based streamline parallel visualization software has been designed and implemented. The software consists of data preprocessing module, thread scheduling and management modules, streamline growing module and graphics rendering module. We use multiple sets of data to verify the correctness and validity of the proposed methods.