鲁棒性P-M方程的快速实现

The Fast Realization of the Robust P-M Model

针对以P-M模型为代表的偏微分方程图像处理计算量大、计算时间长等不足,本文研究了鲁棒性P-M模型,在求解偏微分方程时,为了保证求解精度,提高计算效率,采用了基于CUDA实现的共轭梯度法。实验表明,与传统的基于CPU实现的Thomas快速算法相比,本文实现方法计算所需时间明显减少,而且滤波效果也比显式计算方案明显增强。

Since the image processing using partial differential equation, including the P-M model,has such drawbacks as the huge cumputation and large time computation cost, this paper has made some reserarch on the robust P -M model. When resolving the PDE, this paper use the conjugate gradient method based on CUDA in order to secure the preciseness and increase the computing efficiency. The experiment shows that ,compared with the Thomas fast algorithm using CPU, this method has reduced the computing time apparently and the filtering effort is also better than the one achieved by the explicit numeric computation.