Over the past two decades, flow visualization has played an increasingly important role in industry and engineering areas that involve vector data such as computational fluid simulation (CFD), climate monitoring and simulation and so on. It has become one of the key tools to analyze the flow data and gain interesting information. As the simulation data grow larger and more complex, several challenges need to be overcome. This thesis focuses on performance improvement and more effective seeding strategy for generation of integration-based geometry in flow field. Geometric objects generated by the integration of particles such as streamlines and stream surface are very popular for visualization of flow data, which is easy to be perceived and implemented. However, there are three problems that need to be addressed. First, generation of a large amount of geometric objects for large datasets can consume too much computational resources and cannot be interactively explored. Second the automatic selection of seeds for particles is still a big challenge because of different criteria for different applications. Finally the serious clutter problem in 3D space may inhibit the understanding of flow field. This work improves the performance of particle tracing using parallel computation accelerated by hardware. The algorithm is implemented using OpenCL which can utilize both CPU and graphics processing unit (GPU). Based on the concept of entropy introduced from the information theory, two effective seeding strategies are developed. The entropy value is used to quantize the importance of flow field. The greedy seeding strategy and Monte Carlo method can generate optimized distribution of seed points which assign more points to areas with high entropy values. Three rendering techniques, illumination, animation and halo effect, are implemented to improve the perception of streamlines. Finally a workflow-based flexible visualization system is designed based on the pipeline of VTK. The professional view and Data Gallery view are designed for experienced as well as novice users. Flexible multiple window and multiple view support are also implemented in our system. Five flow datasets are used to demonstrate the effectiveness of techniques in this work.