计算机应用 ›› 2011, Vol. 31 ›› Issue (12): 3334-3336.

• 先进计算 • 上一篇    下一篇

归一化互相关灰度图像匹配的多核信号处理器实现

刘毅飞1,2,张旭明1,丁明跃1   

  1. 1. 华中科技大学 生命科学与技术学院,武汉 430074
    2. 咸宁学院 生物医学工程学院,湖北 咸宁 437100
  • 收稿日期:2011-05-24 修回日期:2011-07-08 发布日期:2011-12-12 出版日期:2011-12-01
  • 通讯作者: 刘毅飞

Multicore digital signal processor implementation of normalized cross-correlation image matching

LIU Yi-fei1,2,ZHANG Xu-ming2,DING Ming-yue2   

  1. 1. College of Biomedical Engineering, Xianning University, Xianning Hubei 437100, China
    2. College of Life Science and Technology, Huazhong University of Science and Technology, Wuhan Hubei 430074, China
  • Received:2011-05-24 Revised:2011-07-08 Online:2011-12-12 Published:2011-12-01
  • Contact: LIU Yi-fei

摘要: 为了满足图像处理对处理器性能的高要求,以基于灰度的归一化互相关(NCC)匹配算法为例,采用高性能、低功耗的多核数字信号处理器(DSP)系统,根据归一化互相关算法中模板图像在源图像中逐个像素搜索并计算相关性的特点,将搜索区域分成六个部分并使TMS320C6472的六个核并行搜索计算这六个区域,并在不同图像存储位置采用不同图像和模板大小实现了多核DSP归一化互相关图像匹配算法。实验结果表明,多核DSP具有作为数字信号处理器的高速信号和图像处理的特点,同时可以根据不同算法通过核间任务分配实现多核并行处理。对于归一化互相关灰度图像匹配算法,TMS320C6472六核DSP和单核DSP比较获得接近单核DSP六倍的性能,对于较大尺寸的图像和PC相比也具有一定的性能加速。

关键词: 多核, 信号处理器, 图像匹配, 归一化互相关

Abstract: In order to meet the high requirements for the performance of processor in image processing, the multicore Digital Signal Processor (DSP) system of high performance and low consumption was adopted. Based on template image in the source image pixel by pixel searching and calculating the relativity in the Normalized Cross-Correlation (NCC) algorithm,the search area was divided into six parts and made each core of six core DSP of TMS320C6472 parallel search and calculate the relativity of the respective area. NCC image matching algorithm was implemented on six core DSP of TMS320C6472 using source images and template images of different sizes which were stored in different memory types. The experimental results show that the multicore DSP inherits such advantages of single DSP as high speed signal and image processing and it can realize multicore parallel processing through inter-core task distribution according to different algorithms. For the NCC image matching algorithm, the performance of six core DSP is almost six times better than a single DSP performance and has improved in comparison to the performance of PC while the image size is relatively large.

Key words: multicore, Digital Signal Processor (DSP), image match, normalized cross correlation