基于特征区域图像自动配准.docVIP

基于特征区域图像自动配准 　　摘 要:为了解决基于特征的图像配准中的特征点的定义和提取问题,提出了一种以特征区域替代特征点的定义和提取方法。该方法应用Moravec算子选择候选特征区域,使用具有旋转不变性的Zernike矩表征该区域的特性;采用二级匹配策略进行特征区域的匹配,即基于自组织映射神经网络的初始匹配及精细匹配;建立图像的配准框架并实现图像的配准。实验结果表明,该方法能有效地提取图像的特征点并能准确地进行特征点的匹配，整个配准过程完全自动进行。 　　关键词:图像配准;特征点;特征区域;Zernike矩;二级匹配策略 　　中图分类号: TP317.4;TP391.41 文献标志码:A 　　? 　　Automatic image registration based on feature region 　　? 　　SHU Xiao?hua??1,2?*?, SHEN Zhen?kang?2 　　?( 　　1.School of Electrical and Information Engineering, Hunan University of Technology, Zhuzhou Hunan 412008, China?;?? 　　2.ATR Laboratory, National University of Defense Technology, Changsha Hunan 410073, China 　　)? 　　Abstract: 　　In order to solve the problem of feature points definition and extraction in image registration based on feature points, an approach was proposed in this paper. Feature region was defined and extracted instead of feature point. Moravec operator was applied to choose the preparatory feature regions, and rotation?invariant Zernike moment was used to characterize the feature regions. Two?step matching strategy was employed for matching the feature regions, i.e. the initial matching was based on self?organizing mapping network and the fine matching. The automatic image registration framework was established and the image registration was realized. The experiments show that this method can effectively extract the image feature points and perform accurate matching of the feature points, the registration process is completely automated. 　　?Key words: 　　image registration; feature point; feature region; Zernike moment; two?step matching strategy 　　? 　　 　　0 引言? 　　与基于像素的图像配准方法相比,基于特征的配准方法因其配准速度较快而得到广泛应用??[1]?。但这类方法需要提取图像的特征点及将特征点进行对应,因此特征点的提取及对应成为这类方法的关键。目前,特征点的定义及提取多采用图像中具有特定意义的点或线（如采用多尺度Harris算子的角点检测??[2]?、尺度不变特征转换(Scale Invariant Feature Transform, SIFT)特征点的检测与匹配??[3]?以及基于结构边缘的特征检测??[4]?等),或采用将特征区域的外部边缘与质心一起作为特征点??[5]?。在这些方法中,特征点的定义与提取均针对图像中的点或线特征。但有些图像中并没有明显的点或线特征,或者其中的点或线特征不容易提取。因此,本文对特征点的定义和选择进行了扩充,将图像中具有特定意义的局部区域作为图像的特征,并据此提