摘要
为提升当前视频去雨算法的去雨效果,同时减少运算时间,在半二次分裂法框架的基础上,提出了一种将梯度先验与卷积神经网络先验相结合的方法进行去雨。在梯度先验的选择上采用了背景层的时间方向梯度和雨滴层的水平方向梯度,同时提出了一种层数参数较少的卷积神经网络作为另一种先验。通过使用软阈值等算法来迭代优化模型函数,获得输出最优解。实验结果表明,方法相较于多个对比方法雨滴去除更为干净且细节保留更好,在峰值信噪比(PSNR)及结构相似度(SSIM)指标上均有所提升,同时在运算时间上具有较大优势。
In order to improve the rain removal effect of the current rain removal algorithm and reduce the calculation time,this paper combines the gradient prior and the Convolutional Neural Network prior to build the rain removal model based on the framework of Half Quadratic Splitting method.In the selection of the gradient prior,the temporal gradient of the background layer and the horizontal gradient of the rain layer are used.At the same time,a Convolutional Neural Network with fewer layers and parameters is proposed as another prior.By using algorithms such as Soft Thresholds to iteratively optimize the model function to obtain the optimal output solution.Compared with multiple methods,this method has cleaner raindrop removal and better detail retention.It has improved Peak Signal-to-Noise Ratio(PSNR)and Structural Similarity(SSIM),and has a great advantage in computing time.
出处
《工业控制计算机》
2021年第8期93-95,共3页
Industrial Control Computer
关键词
视频去雨
方向梯度
卷积神经网络
半二次分裂法
video rain removal
directional gradient
convolutional neural network
half quadratic splitting