摘要
归一化互相关算法是传统的图像匹配算法,针对传统图像匹配算法运算量大、速度慢的缺点,提出了一种基于小波变换的多尺度图像匹配算法。首先在尺度空间的最高一层对低分辨率的子图像进行匹配,然后在匹配结果基础上对高分辨率的图像进行匹配,最终实现全分辨率下的图像匹配。实验结果表明,该算法能够提高图像匹配的精度,减少运算量,满足机器视觉的实时性要求。
Normalized cross-correlation is a traditional image matching algorithm. To the huge amounts of calculation of correlation factor, an image matching algorithm based on wavelet multi-resolution analysis is proposed. Firstly, matching sub images in the lower resolution level. Then this matching result is passed onto the higher resolution level as an initial estimate. Finally, the matching between the higher resolutions images is implemented to the full resolution images. The test shows that the modified algorithm improve considerably the efficiency and accuracy of image matching which meet the real-time request in machine vision system.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第20期4674-4676,共3页
Computer Engineering and Design
关键词
归一化互相关
图像匹配
多尺度分解
小波变换
机器视觉
normalized cross-correlation image matching algorithm
image matching
multi-scale decomposition
wavelet transform
machine vision
