基于多通道融合和组稀疏编码的视频去雪算法

Video Desnowing Algorithm Based on Multi-Channel Fusion and Group Sparse Coding

大雪天气严重降低成像设备所采集视频的能见度,视频去雪算法可以恢复降雪视频的质量.为了去除降雪视频中的雪花,提出一种新的多通道融合和组稀疏编码去雪算法.针对视频的每一个色彩通道中存在的雪花成分进行去除,提出了一种全新的基于低秩矩阵分解的多通道融合背景建模方法,用于恢复干净的背景.为了检测雪花和运动前景,将运动成分中的雪花和运动前景分离以保留运动前景部分,提出了一种基于L0正则化的阈值化方法检测运动物体并分离雪花像素.最后,对被细小雪花遮挡的前景物体采用基于空间适应奇异值阈值的组稀疏编码进行去雪花处理,得到干净的前景.将干净的背景视频和干净的前景视频合成为一段完整的去雪后的视频.

The heavy snow weather severely reduces the visibility of the video captured by the imaging devices,and the video snow removal algorithm can restore the quality of the snow video.In order to remove the snowflakes in the snowfall video,a new method is proposed to remove snow based on multi-channel fusion and group sparse coding.This paper proposed to remove snowflake components in each color channel of the video,and a new multi-channel fusion background modeling method based on low-rank matrix decomposition is proposed to restore a clean background.In order to detect snowflakes and motion foreground,the snowflakes and motion foreground in the motion component are separated to preserve the motion foreground component,and a thresholding method based on the L0 regularization term is proposed to detect moving objects and separate snowflake pixels.A spatially adaptive iterative singular-value thresholding-based group sparsity denoising method is utilized to remove the foreground occluded by small snowflakes to obtain clean foreground.A clean background video and a clean foreground video are combined into a complete snow-free video.