摘要
该文研究了一种利用GPU并行架构的CUDA来完成FDK三维图像重建算法的加速。分析了FDK三维图像重建算法的可并行性特点,设计了适合CUDA的并行方法,分别在算法加权、滤波和反投影3个阶段,实现FDK的加速。经过实验验证,该文提出的方法与算法与CPU单独实现图像重建相比,不仅获得了150倍以上的加速效果,并且两种方式完成的重建图像,质量接近,平均误差小于10^(-4)。由此可得出结论,利用GPU的三维锥束CT图像重建能够得到较满意的结果。
This paper studies a kind of using CUDA GPU parallel architecture to complete the FDK three- dimensional image reconstruction algorithm acceleration. In this paper,it analyzed the FDK algorithm parallelism characteristics,it designed for CUDA parallel method in weighted,filtering algorithm and the projection three phases. It realized the FDK acceleration. Through experiment compared with the algorithm implementation alone by CPU image reconstruction,it not only won more than 150 times speedup,but also got better quality of the image. The average error is less than 10^（-4）. So it concluded that use of GPU 3D cone beam CT image reconstruction can obtain satisfied results.
出处
《工业仪表与自动化装置》
2015年第6期3-7,共5页
Industrial Instrumentation & Automation
基金
中央高校项目(319201300160)
