一种基于频域的运动模糊视频复原方法

A motion blur video restoration method based on frequency domain

视频广泛使用于日常生活、公共安全和空间探索等领域。摄像机抖动和物体运动都会引起采集到的视频发生运动模糊现象。模糊的视频不仅会影响视觉感受,而且会影响对视频中信息的提取和辨识工作,因此需要对模糊视频复原。采集到的运动模糊视频中的模糊具有随机性,对于同一个物体,可能在某些视频帧中是模糊的,但是在另一些视频帧中可能是清晰的。根据这一特征,提出一种基于频域的运动模糊视频复原方法。该方法通过在频域中利用曲波变换融合多个相邻视频帧,获得清晰视频。该方法不需要估计模糊核和迭代求解模型,具有较快的运算速度。曲波变换具有很强的方向敏感性和各向异性特征,有助于模糊视频中边缘信息的复原。实验结果证明,提出的复原方法与现有的复原方法相比具有较好的复原效果。

Videos are widely used in daily life,public safety,space exploration and other fields.Camera shake and object motion can cause motion blur in the captured video.Blur video not only affects the visual experience,but also may affect the extraction and identification of information in the video.Therefore,it is necessary to conduct blur video restoration.The blur of the captured motion blur video is of randomness.For the same object,it may be blurry in some video frames,but it may be clear in other video frames.According to this feature,a motion blur video restoration method based on frequency domain is proposed,in which several adjacent video frames are fused in frequency domain by curvelet transform to obtain clear video.This method does not need to estimate blur kernel and iterate solution model,so it has a fast computation speed.Curvelet transform has characteristics of strong orientation sensitivity and anisotropy,which is helpful to restore the edge information in blur video.The experimental results show that,in comparison with the existing methods,the proposed method has better restoration effectiveness.