提出了一种具有子像素级精度的图像配准方法,并通过迭代反投影算法进行超分辨率重建。介绍了基于3参数模型的图像配准方法并提出了基于4参数模型的图像配准方法,由于没有直接将旋转角度参量引入到运动模型中,有效地避免了原方法中泰勒级数展开时的小角度假设。最后,根据配准算法所得到的子像素级运动信息,用迭代反投影算法进行超分辨率重建。分别对多幅具有较大运动的模拟及真实的低分辨率图像进行实验,结果表明:该配准算法取得了更高的精度,平均平移误差减少了0.026 1 pixel ,平均旋转角度误差减少了0.356 4°;重建图像具有更好的视觉效果,平均PSNR值提高0.75 dB。本文方法可以被广泛地应用于相互间主要存在平移和大角度旋转的多幅低分辨率图像的大倍数超分辨率重建,可满足实际应用要求。

In order to improve the precision of registration algorithm and the quality of reconstructed image, an image registration method with sub-pixel precision is proposed,and super-resolution reconstruction is performed through iterative back projection algorithm. Firstly, an image registration method based on three-parameter model is introduced. Then, an image registration algorithm based on four-parameter model is proposed. Because the rotation angle is not directly included in the movement model, the hypothesis of small rotation angle in Taylor series expanding in original method is avoided effectively. Finally, according to the movement information with sub-pixel precision gained by registration algorithms, super-resolution reconstruction is performed with iterative back projection algorithm. The experiments on simulated and real low-resolution images with big movement are carried out respectively,and the results indicate that proposed image registration algorithm obtains a higher reconstruction precision. The average translation error is reduced by 0.026 1 pixels and the average rotation angle error is decreased by 0.356 4°. The reconstructed image has better visual effect and the average PSNR is increased by 0.75 dB. These results show that proposed method can be used in the large multiple super-resolution reconstruction of low-resolution images with translation and big rotation, and can meet the requirements of practical application.