摘要
针对传统三次卷积插值算法实现遥感图像放大在运算规模、计算速度等方面的不足,结合GPU的高性能计算优势,提出一种基于Dual-GPU(Graphic Processing Unit)的三次卷积插值并行算法(CCPA),即应用GPU的高性能计算技术将传统的三次卷积插值算法进行并行化处理,将图像的像素点个数平均分配给每个线程块,每个线程针对一个像素,线程在GPU中同时执行,以提高其插值效率。实验结果表明,该算法在保持放大后图像质量的同时,速度得到提升,随着图像分辨率的增大,该算法的优势更明显,在分辨率10240*10240的情况下,用GPU处理的速度比CPU提升了97.7%,用双GPU处理的速度是单GPU的2倍,并且在对放大遥感图像的质量和实时性均要求较高如地震、洪水等灾害的情况下,该算法具有实用价值。
The traditional cubic convolution algorithm has to confront with the problems of large operational scale and slow efficiency,when it is used to realize the remote sensing image magnification. In this paper, GPU, as a burgeoning high performance computing technique, was proposed to make parallel processing of the traditional cubic convolution, which we call the Cubic Convolution Parallel Algorithm（CCPA）. This algorithm that divides the pixels points equally to each block, guarantees each pixel point is executed by a thread and threads are executed simultaneously in GPU, impro- ving the interpolation efficiency greatly. The experimental results show that compared with the traditional cubic convo- lution algorithm, this algorithm not only increases the calculation speed, but also achieves high quality image after zoo- ming. Meanwhile, with the growth of image resolution, the advantages of the algorithm become more and more obvious, for instance,to the image of 10240 ~ 10240 resolutions, the speed processed by GPU is 97. 7% higher than that by CPU,and the speed processed by double-GPU is twice than the speed processed by single GPU. Moreover, this algo- rithm also has profound practical value for remote sensing image processing under some emergency situations such as earthquakes, floods and other disasters, with the characteristic of good image quality and real-time mechanism.
出处
《计算机科学》
CSCD
北大核心
2013年第8期24-27,33,共5页
Computer Science
基金
国家重大科技专项高分辨率对地观测系统项目(E0101/1010/01/10
E0104/1112/XT-002)
国家自然科学基金(60973126)
国防科技工业民用专项科研技术研究项目(科工技2010A03A1000)资助
