基于层先验的快速视频去雨算法 下载: 636次
Fast Video Rain Removal Algorithm Based on Layer Priors
上海大学微电子研究与开发中心, 上海 200444
图 & 表
图 1. 背景层的梯度直方图。(a)无雨视频中的图像;(b)垂直方向的梯度;(c)时间方向的梯度
Fig. 1. Gradient histogram of the background layer. (a) Image in the rainless video; (b) gradient in the vertical direction; (c) gradient in the time direction
下载图片 查看原文
图 2. 雨痕层的梯度直方图。(a)雨痕视频中的图像;(b)灰度直方图;(c)水平方向的梯度
Fig. 2. Gradient histogram of the rain trace layer. (a) Image in the rain trace video; (b) gray histogram; (c) gradient in the horizontal direction
下载图片 查看原文
图 3. 不同算法对数据集man的识别结果。(a)输入图像;(b)原始图像;(c)文献[
2]的算法;(d)文献[
3]的算法;(e)文献[
6]的算法;(f)本算法
Fig. 3. Recognition results of different algorithms on the dataset man. (a) Input image; (b) original image; (c) algorithm of Ref. [2]; (d) algorithm of Ref. [3]; (e) algorithm of Ref. [6]; (f) our algorithm
下载图片 查看原文
图 4. 不同算法对数据集trunk的识别结果。(a)输入图像;(b)原始图像;(c)文献[
2]的算法;(d)文献[
3]的算法;(e)文献[
6]的算法;(f)本算法
Fig. 4. Recognition results of different algorithms on the dataset trunk. (a) Input image; (b) original image; (c) algorithm of Ref. [2]; (d) algorithm of Ref. [3]; (e) algorithm of Ref. [6]; (f) our algorithm
下载图片 查看原文
图 5. 不同算法在真实有雨图像上的处理结果。(a)输入图像;(b)文献[
2]的算法;(c)文献[
3]的算法;(d)文献[
6]的算法;(e)本算法
Fig. 5. Processing results of different algorithms on real images with rain. (a) Input image; (b) algorithm of Ref. [2]; (c) algorithm of Ref. [3]; (d) algorithm of Ref. [6]; (e) our algorithm
下载图片 查看原文
表 1本算法的流程
Table1. Flow of our algorithm
Input: rain image, pre-trained GMM model |
---|
Initialization: B=O,R=0, train a lightweight GMM model online | Step1: update M,N by soft threshold algorithm | Step2: update B,R by FTV algorithm | Step3: update ,by approximate optimization algorithm | End: meet the convergence condition Eq. (14) or reach the maximum number of iterations, otherwise go back to step1 and iterate again | Output: background layer B and rain trace layer R |
|
查看原文
表 2实验结果
Table2. Experimental results
Dataset | Algorithm | PSNR /dB | SSIM | Time /s |
---|
Man | Ref. [2] | 37.02 | 0.981 | 137.4 | Ref. [3] | 33.60 | 0.989 | 1332.7 | Ref. [6] | 36.18 | 0.885 | 515.8 | ours | 39.79 | 0.993 | 52.9 | Dataset | Algorithm | PSNR /dB | SSIM | Time /s | Trunk | Ref. [2] | 32.38 | 0.963 | 140.6 | Ref. [3] | 32.86 | 0.962 | 1811.2 | Ref. [6] | 30.32 | 0.938 | 462.2 | ours | 34.57 | 0.976 | 51.9 | Park | Ref. [2] | 34.07 | 0.991 | 101.4 | Ref. [3] | 33.17 | 0.970 | 1501.3 | Ref. [6] | 24.18 | 0.796 | 393.1 | ours | 35.92 | 0.983 | 44.0 | Highway | Ref. [2] | 24.20 | 0.953 | 140.1 | Ref. [3] | 24.12 | 0.936 | 1850.2 | Ref. [6] | 23.78 | 0.868 | 498.7 | ours | 24.81 | 0.958 | 50.2 |
|
查看原文
潘登, 严利民. 基于层先验的快速视频去雨算法[J]. 激光与光电子学进展, 2021, 58(6): 0610013. Pan Deng, Yan Limin. Fast Video Rain Removal Algorithm Based on Layer Priors[J]. Laser & Optoelectronics Progress, 2021, 58(6): 0610013.