No-Reference Image Quality Assessment Algorithm Based on Semi-Supervised Learning

Fig. 6. Scatter plot of prediction results of the proposed algorithm on each dataset. (a) LIVE dataset; (b) CSIQ dataset; (c) TID2008 dataset; (d) TID2013 dataset

Download full size

Table 1. Quality score regression network
View table
Table 1. Quality score regression network
Name Layer
FC1 （1000，1024）
FC2 （1024，512）
FC3 （512，1）

Table 2. Comparison of SVR model prediction and five full-reference image quality evaluation algorithms
View table
Table 2. Comparison of SVR model prediction and five full-reference image quality evaluation algorithms
Dataset Parameter SSIM FSIMc GMSD VSI SPSIM Proposed algorithm
LIVE SROCC 0.948 0.965 0.960 0.952 0.962 0.986
PLCC 0.945 0.965 0.960 0.943 0.960 0.963

Table 3. Comparison of image quality assessment on each dataset

View table

Table 3. Comparison of image quality assessment on each dataset

Dataset	Number of referenced images	Number of distorted images	Evaluation parameter	Parameter range
LIVE	29	779	DMOS	［0，100］
CSIQ	30	866	DMOS	［0，1］
TID2008	25	1700	MOS	［0，9］
TID2013	25	3000	MOS	［0，9］
Challenge		1169	MOS	［0，100］
KonIQ-10k		10073	MOS	［0，5］

Table 4. Performance comparison of different algorithms on artificial simulation datasets

View table

Table 4. Performance comparison of different algorithms on artificial simulation datasets

Dataset	Parameter	MLV	A-D	RISE	DIQA	DB-CNN	Proposed algorithm
LIVE	SROCC	0.931	0.968	0.949	0.970	0.968	0.983
LIVE	PLCC	0.943	0.973	0.962	0.972	0.971	0.979
CSIQ	SROCC	0.925		0.922	0.843	0.946	0.957
CSIQ	PLCC	0.949		0.942	0.868	0.959	0.953
TID2008	SROCC	0.855	0.865				0.928
TID2008	PLCC	0.858	0.869				0.939
TID2013	SROCC	0.879	0.812	0.934	0.844	0.816	0.926
TID2013	PLCC	0.883	0.845	0.942	0.880	0.865	0.934

Table 5. Performance comparison of different algorithms on natural distortion datasets
View table
Table 5. Performance comparison of different algorithms on natural distortion datasets
Algorithm Challenge KonIQ-10k
SROCC PLCC SROCC PLCC
HOSA 0.640 0.678 0.671 0.694
WaDIQaM-NR 0.671 0.680 0.797 0.805
PQR 0.857 0.882 0.880 0.884
SFA 0.812 0.833 0.856 0.872
Proposed algorithm 0.871 0.863 0.893 0.845

Table 6. Ablation experimental results
View table
Table 6. Ablation experimental results
Module LIVE Challenge
SROCC PLCC SROCC PLCC
RepVGG+ReLU 0.967 0.957 0.841 0.822
RepVGG+P-ReLU 0.969 0.962 0.845 0.835
RepVGG+ReLU+self 0.981 0.975 0.869 0.857
RepVGG+P-ReLU+
self（ours）
0.983 0.979 0.871 0.863

Tools

Get Citation

Copy Citation Text

Xiangdong Jin, Qingbing Sang. No-Reference Image Quality Assessment Algorithm Based on Semi-Supervised Learning[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0410023

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Jan. 14, 2022

Accepted: Mar. 30, 2022

Published Online: Feb. 14, 2023

The Author Email: Sang Qingbing (sangqb@163.com)

DOI:10.3788/LOP220543

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Quality score regression network

Table 1. Quality score regression network

Table 2. Comparison of SVR model prediction and five full-reference image quality evaluation algorithms

Table 2. Comparison of SVR model prediction and five full-reference image quality evaluation algorithms

Table 3. Comparison of image quality assessment on each dataset

Table 3. Comparison of image quality assessment on each dataset

Table 4. Performance comparison of different algorithms on artificial simulation datasets

Table 4. Performance comparison of different algorithms on artificial simulation datasets

Table 5. Performance comparison of different algorithms on natural distortion datasets

Table 5. Performance comparison of different algorithms on natural distortion datasets

Table 6. Ablation experimental results

Table 6. Ablation experimental results