摘要
常规的图像相关匹配算法通常只具有平移不变性 ,而不具有尺度和旋转不变性 ,这就给存在尺度和旋转变化图像间的匹配问题设置了障碍。针对这一问题 ,提出了一种基于图像对数极坐标变换的相关匹配算法 ,并融入了高斯金字塔多分辨率机制以提高算法的效率。由于对数极坐标变换具有尺度和旋转不变性 ,因此该方法能够较好地克服传统图像匹配算法的缺陷 ,不仅可以完成存在更一般运动的图像间的匹配 。
Traditional image correlation matching algorithm works under a pure translation model and would face difficulty when dealing with a more general motion model where either scaling or rotation of the matched images is present. Accordingly, a novel multi-resolution correlation matching algorithm based on log-polar image transform is proposed. First, the traditional correlation matching algorithm based on Gaussian multi-resolution scheme is proposed and inherent defects of the method are analyzed. Afterwards, the log-polar image transform is introduced and its invariance properties with respect to scale and rotation changes are illustrated. Finally, the correlation matching algorithm based on log-polar image transform is implemented. Simulation results show that due to the invariance properties of log-polar image transform, correlation matching between images with scale and rotation changes can be well achieved and the magnitudes of changes can also be accurately estimated through further correlation matching in the log-polar space.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2004年第5期653-656,共4页
Journal of Northwestern Polytechnical University
关键词
相关匹配
对数极坐标变换
多分辨率
correlation matching, log-polar transform, multi-resolution
