摘要
针对已知指纹图像增强系统中指纹局部频谱特性估计的计算负荷大,基于CPU实现后执行时间较长的问题,提出了一种基于GPU的CUDA并行实现方法来提高运算速度.利用不同CUDA线程块来并行处理不同的局部指纹图像,同时线程块中的多线程对相应局部指纹的频谱特性估计进行并行优化,从而完成整个指纹图像的局部频谱特性估计的优化.通过对FVC2000数据库中大量的指纹图像进行测试,得到相应的执行时间并与其基于CPU实现的执行时间相比较.结果显示:通过该基于GPU的CUDA并行实现方法,局部指纹频谱特性估计的执行时间显著减少,从而可以提高已知指纹图像增强系统的运算速度.
Aiming at the long execution time of the local spectrum feature estimation based on CPU in the existing fingerprint enhancement system,an implementation method based on parallel GPU programming is proposed to reduce the execution time.Different CUDA blocks are invoked to process different local fingerprints in parallel.And the multi-threads in each CUDA block are fully utilized to optimize the local spectrum feature estimation in parallel.Large number of fingerprint images in the FVC2000 database are tested and the results show that compared with the CPU-based implementation method,the execution time of the local fingerprint spectrum feature estimation is significantly reduced using GPU-based implementation method.It indicates that the GPU-based implementation method can be applied to the fingerprint enhancement system to improve the computing speed which will make it more applicable.
作者
南余荣
王福良
NAN Yurong;WANG Fuliang(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《浙江工业大学学报》
CAS
北大核心
2018年第5期492-495,共4页
Journal of Zhejiang University of Technology
关键词
指纹
图像增强
CUDA编程
GPU
并行运算
fingerprint
image enhancement
CUDA programming
GPU
parallel programming
作者简介
南余荣(1966—),男,浙江乐清人,教授,博士,研究方向为电力传动及其自动化,E-mail:nyr@zjut.edu.cn.
