基于CUDA的拉普拉斯边缘检测算法

拉普拉斯边缘检测算法常用于去除CCD天文图像中的宇宙射线噪声，但其串行算法计算复杂度较高。为此，分析拉普拉斯边缘检测算法的并行性，在统一计算设备架构(CUDA)并行编程环境下，提出一种基于CUDA的拉普拉斯边缘检测图形处理单元(GPU)并行算法。分割天文图像得到多幅子图，根据GPU的硬件配置设定Block和Grid的大小，将子图依次传输到显卡进行并行计算，传回主存后拼接得到完整的图像输出。实验结果表明，图像尺寸越大，该并行算法与串行算法相比具有的速度优势越大，可获得10倍以上的加速比。

Laplacian Edge Detection Algorithm Based on CUDA

Laplacian edge detection algorithm is widely used in the removal of cosmic ray noise in the CCD astronomical images,but it has higher computation complexity of single CPU.To solve this problem,this paper proposes a parallel Laplacian edge detection algorithm with Graphic Processor Unit(GPU) based on Compute Unified Device Architecture(CUDA) by analyzing the parallelism of Laplacian edge detection.The main algorithm running on the main CPU is responsible for split the astronomical image into some subgraphs.Then it sets Block and Grid size according to the GPU hardware configuration,and transfers the subgraph to graphics card for parallel computing.Finally it retrieve the processed subgraph to main memory and joining together to get complete image output.Experimental results show that,with the image size increases,the speed advantage of the parallel algorithm is greater than the serial algorithm,and it obtains more than ten times speedup measured.