Feature Distribution Based Visualization Research In 3D Flow Field

The flow fields are widely used by researchers in the fields of scientific computation and engineering analysis.It is the transparence of the flow fields that makes them difficult to present the data from measurement and computation to users in a specific way with graph images.To solve such a problem,researchers add nontransparent outer materials to the transparent flow fields,and make them become the important tracks to reflect the regulation of the flow movement.That is how the flow field visualization works.The visualization of flow field is always a hotspot in vector field visualization,and it has wide development prospect and application.Because the streamline-based visualization is the most widely used geometrical method that describes the structures of the flow fields and behaviors in the numerous methods,and its results are clear and simplicity,and it is extensively applied and researched.One significant goal of the streamline-based visualization is to reveal as many the interesting features as possible to enhance the perception of users about the data,and represent full effective and correct visualization results.Theoretically,provided that enough dense streamlines have been placed in the flow field,the significant information would not be missed.In the practice,however,a large amount of streamlines in the 3D flow field would influence the visual impressions.Too many streamlines not only make the 3D flow field clutter,but also lead to occlusion and cover up the considerable features.Therefore,how to place flow lines as few as possible in the 3D flow field without losing important information is an ultimate goal in flow visualization.The paper does visualization research on the 3D flow field data sets,and utilizes geometrical feature distribution of streamlines to describe the flow fields,and proposes an efficient streamlines feature distribution based method.We select the curvature,torsion,tortuosity and velocity direction entropy as the feature descriptors,and construct the 2D histograms of the feature descriptor to describe the shape feature of the streamlines.The 2D histograms have the spatial information,which can remedy the measurement disadvantage caused by the 1D histograms that lose the spatial information.We measure the similarities of the histograms with the Earth mover's distance(EMD),and avoid taking too much time in computing the closest points with iteration by iterative closest points algorithm.We apply k-means method to cluster the streamlines,remove the same percentage of the redundant streamlines according to the requests of streamline density from users.We adopt the CUDA technique to speed up the computation of visualization.The experimental results show that the method can represents the main structure features correctly,and enhance the readability of the visualization results.