摘要
为了从由水面波动引起严重失真的视频中快速恢复出真实的水下场景图像,提出了一种结合序列图像配准和最优图像块选择的复原算法。首先通过一种迭代的序列图像配准算法消除视频帧中严重的几何畸变,并获得任意时刻水表面的三维形状,然后利用最优图像块选择算法从校正后的图像序列中合成出无失真的水下场景图像。实验结果表明,与主流的图像配准结合稀疏噪声去除的方法相比,算法能够获得更加准确清晰的视觉效果,同时具有更高的计算效率。
With the aim to recover the planar underwater scenes from a video sequence severely distorted by water waves, a reconstruction framework which integrated robust registration with lucky image approaches was proposed. At first, an iterative robust registration algorithm was used to eliminate most geometric deformations and recover the water surface. Then the best image patches selected from the corrected video frames were stitched together. With the experimental results, it is found that, in terms of restoration accuracy, visual effects and computational efficiency, the suggested algorithm always significantly outperforms better than the method employing robust registration and sparse noise elimination, leading to state-of-the-art performance on the task of underwater image restoration from video sequences.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2012年第1期188-191,196,共5页
Journal of System Simulation
关键词
流体表面重建
水下场景复原
图像配准
最优图像块选择
water surface reconstruction
underwater image restoration
image registration
lucky image patch