Image Registration-oriented Research Of Subpixel Motion Estimation Algorithm

The basic task of motion estimation is to detect the motion information between the image frames in the image sequence and functions as an essential research field in the computerized vision. The motion estimation is the essential technology of the image registration. Essentially speaking, the image registration can be deemed as an estimation and transform of model parameters based on a certain motion transform model. The high-accuracy image registration is the basis of such applications as the time variation detection, 3D matching, motion analysis and image sequence analysis and has gradually become a topic of general interest in the current image processing technology research. The subpixel motion estimation is one of the premises and key technology to achieve high-accuracy motion registration.This thesis is conducted on the subpixel motion estimation technology of the image registration and, on the basis of reviewing the algorithms of the traditional integer pixel motion estimation and the current subpixel motion estimation, the research and realization are conducted on the subpixel motion estimation algorithm from the frequency register and the space register. On the motion estimation with the revolving and horizontal movement concurrently, it is firstly simplified as estimating the revolving angle and compensating the revolving angle and then estimating the horizontal parameter estimation, while the estimated angle accuracy of both Hough transform and Log-polar Coordinate; the phase correlation method of the frequency register motion estimation is extended to the subpixel pole and the analytical method is used to get the motion parameter. In addition a method combining the phase correlation and the template matching is proposed herein, which inherits the characteristics of the template matching method as being simple and easy to realize and reaches the subpixel grade accuracy and effectively reduces the calculation volume. An improved subpixel motion estimation algorithm based on the characteristics is proposed for the case of noise potentially existing in the actual application. This algorithm selects points less affected by noise to perform the motion parameter estimation. According to the experiment results, this algorithm not only reaches the subpixel grade accuracy, but achieves good robustness to the influence of noise. In order to further verify the performance of the algorithm, it is applied to the image mosaicing and super-resolution reconstruction and achieves good experiment results, so as to improve the validity of the method.