Parallel computing of texture-optimisation-based flow visualisation on CUDA.

Flow visualisation effectively visualises flow fields with moving textures by vividly capturing the flow field properties through varying texture appearances. Texture-optimisation-based (TOB) flow visualisation can produce excellent visualisation results of flow fields. However, TOB flow visualisation is a slow process with a huge amount of time-consuming computation of nearest neighbour searching and thus is difficult to be applied to dynamic flow field visualisation. In this paper, we propose an optimal acceleration scheme for speed and quality for searching the approximate nearest neighbour by comparing and analysing three techniques to accelerate the computation of the nearest neighbour. We achieve the parallel computation of TOB flow visualisation algorithm based on CUDA implementation on graphics processing unit (GPU). Most time-consuming computations are performed in parallel on GPU, which yields high performance. Experimental results show that our TOB flow visualisation generates results with fast synthesis speed and high synthesis quality. This method can visualise not only static flow fields but also time-varying or dynamic flow fields.