The Study Of Automatic Stream Surface Construction Based On Clustering

Surface-based flow visualization is an important branch of Scientific Visualization,and plays an important role in the field of aerospace.However,classical methods of stream-surface visualization require users to place seed lines manually,which means it's very difficult for users to place seed lines at interesting domain of flow field.A Cluster-based method which can place stream-surfaces automatically is presented in this paper.This method composes of three parts,clustering flow data by their flow behavior,placing seed lines in each cluster,and constructing stream-surface.First,the flow field is partitioned according to the similarity of the velocity gradient,curvature and position.This task is completed by a modified K-means algorithm.The modified K-means algorithm picks original cluster centers in a different way which makes those centers have low similarity with each other.The modified K-means algorithm takes less time for iteration and runs faster.And we can also choose appropriate Cluster Number through Silhouette Method.After partition of flow field by clustering,one seed line is placed in each partition which starts from the center of the partition and grows along the direction of the velocity curvature until it arrives the boundary of its partition.And we find seed lines constructed through this way represents the flow field in a better way and present less occlusion.Finally,stream-surfaces are constructed with the iterative advancing front method starting from each seed line with Hermite interpolation.Traditional advancing front method constructs stream surface through recursion way,which not only takes more stack source,but also need more computation.We design an iterative method which takes less stack source and much less running time.We use Hermite interpolation when a new point is needed instead of insert midpoint directly and this can guarantee the generated stream-surface be C4-continuous.We developed a Paraview plugin based on the above methods,and analyzed several flow field data with this plugin.Our experiments show that stream surfaces generated with our approach can present the main flow structure in most cases.The generated seed lines can also be taken as reference when users want to place seed lines manually for special representation.