基于Fourier-Mellin变换的图像配准方法及应用拓展

从两个方面拓展了基于Fourier—Mellin变换的图像配准方法的应用范围．首先是全景图像的拼接．不同于传统的方法，该方法不需要准确控制相机的运动，小需要知道相机的焦距等内部参数．也不需要检测图像特征，在配准精度要求不是很高的情况下，直接生成的全景图像可以满足很多实际应用的需要；同时，实验也表明，该方法应用于弱透视图像的配准．也具有很好的配准效果．另一个拓展是图像曲线的匹配．传统的曲线匹配方法一般通过曲线特征点（如角点、曲率极值点等）之间的对应求得曲线间的变换参数．一种新的思想是先将图像曲线转化为二值图像，然后应用Fourier—Mellin变换对这些二值图像进行配准，从而达到对两条曲线的匹配．大量实验表明，该方法对射影畸变不是十分显著且摄像机为一般运动下获得的图像之间的配准问题（如手持数码相机获取的图像之间的配准问题）均能取得比较好的配准效果．

An Image Registration Technique Based on Fourier-Mellin Transform and Its Extended Applications

The image registration technique based on Fourier-Mellin transform can be used to register images which are misaligned due to rotation,scaling and translation and find its applications in many different fields thanks to its high accuracy,robustness and low computational cost.In this paper,the technique is extended to two new application fields.The first one is panoramic mosaics.Unlike conventional methods,this technique is capable of successfully building a coarse full view of a large scene without either requiring special hardware to control camera motion or knowing camera's focus length,or detecting image features and their correspondences.The other extended application is of curve matching.In the most traditional curve matching methods,the correspondence of curve features,such as corners,extrema of the curvature etc,should be at first established,and then the matching parameters are computed.Here a new approach is proposed,where the curves matched are at first converted into binary images and then the matching of these binary images is carried out by the Fourier-Mellin transform based registration technique.Numerous experiments show that for most of images captured by a hand-held camera,if the projective distortions are not too severe,the registration results are satisfactory.