Imaging through a time-varying water surface exhibits severe non-rigid geometric distortions and motion blur. Theoretically, although the water surface possesses smoothness and temporal periodicity, random fluctuations are inevitable in an actual video sequence. Meanwhile, considering the distribution of information, the image structure contributes more to the restoration. In this paper, a new two-stage restoration method for distorted underwater video sequences is presented. During the first stage, salient feature points, which are selected through multiple methods, are tracked across the frames, and the motion fields at all pixels are estimated using a compressive sensing solver to remove the periodic distortions. During the second stage, the combination of a guided filter algorithm and an image registration method is applied to remove the structural-information-oriented residual distortions. Finally, the experiment results show that the method outperforms other state-of-the-art approaches in terms of the recovery effect and time.

通过随时间变化的水面成像会表现出严重的非刚性几何变形和运动模糊。从理论上讲，尽管水面具有平滑度和时间周期性，但在实际的视频序列中不可避免地会出现随机波动。同时，考虑到信息的分布，图像结构对恢复的贡献更大。本文提出了一种新的两阶段失真水下视频序列恢复方法。在第一阶段，通过帧跟踪通过多种方法选择的显着特征点，并使用压缩感测解算器估算所有像素的运动场，以消除周期性失真。在第二阶段 结合了导向滤波算法和图像配准方法，消除了面向结构信息的残余畸变。最后，实验结果表明，该方法在恢复效果和时间上均优于其他最新方法。