Automatic Image Registration Based on Matching of Feature Points in Broad Sense
基于广义特征点匹配的全自动图像配准

Feature points in images are commonly used for image registration. Feature points can be classified as in narrow sense and in broad sense. Feature Points in Broad Sense (FPBS) can be defined in different ways. A new definition of FPBS is proposed that is reasonable for image registration. The FPBS can be detected automatically by the use of multi-scale wavelet transform and a few additional restrictions that control the complexity and nonperiodicity of local regions. After feature point sets are extracted separately from the two images under consideration, the relation between them is then established by a two-stage matching algorithm. The registration transform is found by minimizing the Root Mean Square Error (RMSE) of the control points. An iterative optimization mechanism is used to refine the registration. Several experimental results of image registration can illustrate the performance of the method.