基于CPU-GPU混合编程的显微镜图像实时拼接  

Real Time Stitching of Microscope Images Based on CPU-GPU Hybrid Programming Method

作　　者：吴为民[1] 刘新[1] 李伙钦 江先伟[1] 杨华 WU Weimin;LIU Xin;LI Huoqin;JIANG Xianwei;YANG Hua(School of Information and Intelligent Transportation,Fujian Chuanzheng Communications College,Fuzhou 350007,China;Fuzhou Yifan Vision Technology Co.Ltd.,Fuzhou 350007,China)

机构地区：[1]福建船政交通职业学院信息与智慧交通学院,福州350007 [2]福州一方威视科技有限公司,福州350007

出　　处：《重庆科技学院学报（自然科学版）》2023年第3期67-74,共8页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：福建省中青年教师教育科研项目“基于特征融合的图像显著性检测研究”(JAT160715),“基于多模态数据融合的行人重识别研究”(JAT210704);2020年校级横向课题“数码显微镜数字图像拼接算法开发”(CZJTXXXY2020007);2022年校级科教发展基金项目“基于语义协同的交通视觉显著性检测”(20220209)。

摘　　要：随着电子显微镜图像的分辨率越来越高,图像拼接的计算量也越来越大,实时拼接的流畅效果对计算速度提出了很高的要求。利用NVIDIA的GPU并行编程框架CUDA,将拼接过程中耗时较长的图像特征点检测和图像拷贝部分迁移到GPU上进行并行计算,CPU则负责逻辑控制部分的计算,提高了整体的运算效率。实验结果表明,CPU-GPU混合编程模式有效地缩短了显微镜图像拼接时间,提高了拼接的流畅度和实时性。With the increasing resolution of electron microscopy images,the computational complexity of image stitching is also increasing,and real-time smooth effects require high computational speed.In this study NVIDIA′s GPU parallel programming framework CUDA is utilized to migrate the time-consuming image feature point detection and image copy parts during the stitching process to the GPU for parallel computation.The CPU is responsible for the logical control part of the computation,improving the overall computational efficiency.The experimental results show that the hybrid programming mode of CPU and GPU effectively shortens the time of microscopic image stitching and improves the smoothness and real-time performance of stitching.

关 键 词：电子显微镜 实时拼接 并行计算 CPU-GPU混合编程 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]