Image will be seriously distorted when camera captures the scene through water surface. This will cause the captured image to lose its readability. Although several previous methods have been proposed to resolve this problem, the methods still need improvement. In this paper, we propose a reconstruction method to restore the distorted images based on a twice registration algorithm. Several calculation methods will be applied to update reference image in different registration phase. Firstly, the method uses the mean image of the image sequence as the reference image for the first registration. Due to the reference image may be blurred, a deblurring step has been applied in this paper. To register the distorted image sequence, a non-rigid registration method has been employed based on B-spline method. In addition, a patch search method also is proposed to update the reference image in the second registration because reference image is the key to improve the accuracy of image registration. It is designed to obtain the most similar patches from the image sequence. Compared with the previous methods using the existing database and generated database, the results show that our method performs better than the state-of-the-art methods. [ABSTRACT FROM AUTHOR]