Improved Convolutional Rain Removal Algorithm for Single Image Based on Graph Network

Jinxiang Zhou; Zhiwei Li; Huowang Qiu; Yuanhong Ren; Wuneng Zhou

doi:10.3788/LOP213091

Laser & Optoelectronics Progress, Volume. 60, Issue 4, 0410016(2023)

Improved Convolutional Rain Removal Algorithm for Single Image Based on Graph Network

Jinxiang Zhou^1,1、">, Zhiwei Li^1,1,2、">*, Huowang Qiu^1,1、">, Yuanhong Ren^2,2、">, and Wuneng Zhou^2,2、">

Author Affiliations

¹College of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201406, China

²College of Information Science and Technology, Donghua University, Shanghai 201620, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(7)

Fig. 1. Flowchart of the proposed algorithm

Download full size

Fig. 2. Schematic of the unfolding process of convolution kernel

Download full size

Fig. 3. Visual comparison of several latest algorithms before and after improvement

Download full size

Table 1. Performance comparison results of several latest algorithms on four public test datasets

View table

Table 1. Performance comparison results of several latest algorithms on four public test datasets

Algorithm	PSNR/SSIM
Algorithm	Rain100L	Rain100H	RainLight	RainHeavy
DDN	32.16/0.936	21.92/0.764	31.66/0.922	22.03/0.777
DDN+	33.45/0.944	24.02/0.816	32.91/0.937	24.22/0.825
PReNet	37.48/0.979	29.45/0.905	37.93/0.983	29.36/0.903
PReNet+	39.21/0.991	30.55/0.921	39.60/0.993	30.41/0.919
Multi-scale non-local	36.01/0.965	28.63/0.884	36.39/0.968	28.47/0.883
Multi-scale non-local+	37.88/0.984	29.98/0.916	37.93/0.982	29.77/0.914
BRN	38.16/0.982	30.47/0.918	38.86/0.986	30.27/0.917
BRN+	39.55/0.993	31.29/0.934	40.08/0.996	31.11/0.925

Table 2. Comparison results of computing time and GPU memory of several latest algorithms before and after improvement
View table
Table 2. Comparison results of computing time and GPU memory of several latest algorithms before and after improvement
Algorithm Computing Time /s GPU memory /MB
DDN 0.014 1242
DDN+ 0.016 1305
PReNet 0.058 1897
PReNet+ 0.062 1994
Multi-scale non-local 0.104 4957
Multi-scale non-local+ 0.111 5135
BRN 0.089 2587
BRN+ 0.099 2643

Table 3. Comparison results of performance of AODNet
View table
Table 3. Comparison results of performance of AODNet
Parameter AODNet AODNet+
PSNR 19.6954 20.1430
SSIM 0.8478 0.8539

Table 4. Comparison results of computing time and GPU memory of AODNet before and after improvement
View table
Table 4. Comparison results of computing time and GPU memory of AODNet before and after improvement
Parameter AODNet AODNet+
Time /s 0.0087 0.0095
GPU memory /MB 587 614

Tools

Get Citation

Copy Citation Text

Jinxiang Zhou, Zhiwei Li, Huowang Qiu, Yuanhong Ren, Wuneng Zhou. Improved Convolutional Rain Removal Algorithm for Single Image Based on Graph Network[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0410016

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Nov. 29, 2021

Accepted: Jan. 5, 2022

Published Online: Feb. 14, 2023

The Author Email: Li Zhiwei (zhiwei.li@sues.edu.cn)

DOI:10.3788/LOP213091

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Performance comparison results of several latest algorithms on four public test datasets

Table 1. Performance comparison results of several latest algorithms on four public test datasets

Table 2. Comparison results of computing time and GPU memory of several latest algorithms before and after improvement

Table 2. Comparison results of computing time and GPU memory of several latest algorithms before and after improvement

Table 3. Comparison results of performance of AODNet

Table 3. Comparison results of performance of AODNet

Table 4. Comparison results of computing time and GPU memory of AODNet before and after improvement

Table 4. Comparison results of computing time and GPU memory of AODNet before and after improvement