基于全变分正则和L2,1范数的视频去雨张量模型

提出了一种基于全变分正则与$L_{2，1}$范数的视频去雨张量模型用于解决雨线遮挡问题。首先，对雨线成分与视频背景先验信息进行预处理，获取相应正则化条件的构建依据以增强各部分稀疏性，便于促进雨线分离。其次，考虑到视频图像存在不规则动态对象，引入全变分正则项来抑制背景强度变化，缓解雨线的误判现象。采用交替方向乘子法（alternating direction method of multipliers，ADMM）可以有效地对所提出的张量模型进行求解，并在合成数据与真实数据集上开展大量实验。结果表明，所提方法在动态背景情况下有效去除视频图像雨线的同时，保留了更多背景细节信息。与相关先进方法相比，所提方法在峰值信噪比、结构相似性和残差三种综合性能量化指标上均具有较大的优势。

Tensor Model Based on Total Variation Regularized and L2,1 Norm for Video Rain Streaks Removal

This paper proposes a tensor model for video rain streaks removal based on total variational regularization and L_2,1 norm to solve rain streaks shielding. Firstly, the prior information of rain streaks component and the video background are preprocessed to obtain the corresponding regularization condition, so as to enhance the sparsity of each part and facilitate the separation of rain streaks. Then, considering the existence of irregular dynamic objects in the video, a total variational regularization term is introduced to suppress the variation of background intensity and alleviate the misjudgment of rain streaks. The alternating direction method of multipliers (ADMM) can be used to effectively solve the proposed tensor model, and carried out a large number of experiments on the synthetic and real datasets. Experimental results show that the proposed method under dynamic background can effectively remove the video rain streaks and retain more background details simultaneously. Compared with other relevant methods, the proposed method has great advantages in three comprehensive quantitative performance measures of peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and residual error (RES).