No-Reference Stereoscopic Image Quality Assessment Based on Binocular Neuron Response

Mengmeng Ye; Jinbin Hu; Xuejin Wang; Feng Shao

doi:10.3788/LOP202158.2410007

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 24 >Page 2410007 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 24, 2410007 (2021)

No-Reference Stereoscopic Image Quality Assessment Based on Binocular Neuron Response

Mengmeng Ye, Jinbin Hu, Xuejin Wang, and Feng Shao^*

Author Affiliations

Faculty of Information Science and Engineering, Ningbo University, Ningbo, Zhejiang 315211, China

show less

DOI: 10.3788/LOP202158.2410007 Cite this Article Set citation alerts

Mengmeng Ye, Jinbin Hu, Xuejin Wang, Feng Shao. No-Reference Stereoscopic Image Quality Assessment Based on Binocular Neuron Response[J]. Laser & Optoelectronics Progress, 2021, 58(24): 2410007 Copy Citation Text

show less

Fig. 1. Diagram of binocular energy model

Download full size

Fig. 2. Diagram of stereoscopic image quality assessment model

Download full size

Fig. 3. Five types of V1 neuron response images

Download full size

Feature	Feature description	Symbolic representation
DCT domain feature	10% and 100% pooled shape parameter	f₁ and f₂
	10% and 100% pooled coefficient of frequency variation	f₃ and f₄
	10% and 100% pooled energy sub-band ratio	f₅ and f₆
	10% and 100% pooled orientation feature	f₇ and f₈
Spatial feature	(α,σ²) from GGD fit of MSCN coefficients	f₉ and f₁₀
Spatial feature	(η,υ, $σ_{l}^{2}$ , $σ_{r}^{2}$ ) from AGGD fit of four adjacent MSCN coefficients	f₁₁-f₂₆

Table 1. Summary of extracted features

Method	Type	LIVE-I			LIVE-II
Method	Type	PLCC	SROCC	RMSE	PLCC	SROCC	RMSE
FI-PSNR	FR	0.865	0.856	8.242	0.658	0.638	8.496
FI-SSIM	FR	0.870	0.861	8.087	0.684	0.680	8.230
Method in Ref.[8]	FR	0.925	0.922	-	0.759	0.745	-
Method in Ref. [9]	FR	0.889	0.877	7.519	0.770	0.751	7.204
Method in Ref. [10]	FR	0.918	0.909	6.501	0.907	0.901	4.766
Method in Ref. [3]	NR	0.917	0.911	5.864	0.737	0.701	7.665
Method in Ref. [4]	NR	0.910	0.901	6.794	0.750	0.701	7.198
Method in Ref. [13]	NR	0.922	0.903	6.258	0.913	0.905	4.657
Method in Ref. [15]	NR	0.891	0.885	-	0.784	0.805	-
Method in Ref. [22]	NR	0.938	0.868	-	0.851	0.831	-
Proposed method	NR	0.938	0.927	5.583	0.937	0.931	3.901

Table 2. Performance comparison among different assessment metrics on LIVE-I and LIVE-II

Method	LIVE-I					LIVE-II
Method	JPEG	JP2K	WN	GB	FF	JPEG	JP2K	WN	GB	FF
FI-PSNR	0.207	0.839	0.928	0.935	0.658	0.613	0.719	0.907	0.711	0.701
FI-SSIM	0.241	0.822	0.928	0.879	0.687	0.564	0.700	0.909	0.739	0.735
Method in Ref. [8]	0.615	0.875	0.943	0.938	0.781	0.720	0.848	0.846	0.801	0.851
Method in Ref. [9]	0.347	0.819	0.908	0.918	0.653	0.846	0.804	0.939	0.884	0.874
Method in Ref. [10]	0.440	0.865	0.937	0.924	0.758	0.840	0.833	0.955	0.910	0.889
Method in Ref. [3]	0.699	0.890	0.899	0.922	0.649	0.566	0.704	0.459	0.896	0.711
Method in Ref. [4]	0.570	0.812	0.940	0.878	0.784	0.605	0.695	0.440	0.860	0.683
Method in Ref. [13]	0.603	0.838	0.906	0.791	0.679	0.818	0.845	0.946	0.903	0.899
Method in Ref. [15]	0.693	-	0.899	0.853	-	0.622	-	0.803	0.713	-
Method in Ref. [22]	0.633	-	0.920	0.903	-	0.788	-	0.929	0.909	-
Proposed method	0.742	0.891	0.920	0.867	0.737	0.847	0.850	0.942	0.914	0.903

Table 3. SROCC results of different assessment metrics on LIVE-I and LIVE-II for different distortion types

Method	Type	NBU-MDSID I			NBU-MDSID II
Method	Type	PLCC	SROCC	RMSE	PLCC	SROCC	RMSE
Method in Ref. [8]	FR	0.919	0.905	3.687	0.802	0.862	7.212
Method in Ref. [9]	FR	0.856	0.834	4.943	0.820	0.780	7.110
Method in Ref. [10]	FR	0.885	0.877	4.385	0.763	0.749	7.560
Method in Ref. [3]	NR	0.937	0.930	3.244	0.835	0.800	6.589
Method in Ref. [4]	NR	0.920	0.900	3.701	0.824	0.799	6.827
Method in Ref. [14]	NR	0.934	0.920	3.348	0.791	0.763	7.180
Method in Ref. [15]	NR	0.878	0.882	4.570	0.606	0.627	9.586
Method in Ref. [16]	NR	0.916	0.922	3.836	0.785	0.765	7.442
Method in Ref. [30]	NR	0.938	0.926	3.062	-	-	-
Method in Ref. [22]	NR	0.940	0.936	3.804	0.845	0.819	7.020
Proposed method	NR	0.963	0.952	3.887	0.859	0.851	6.110

Table 4. Performance comparison among different assessment metrics on NBU-MDSID I and NBU-MDSID II

Database	Index	Neuron response
Database	Index	ODF TE	ODF TI	ODF ODD	RPC TE	RPC ODD	All
LIVE-I	PLCC	0.944	0.942	0.944	0.943	0.946	0.938
LIVE-I	SROCC	0.934	0.925	0.933	0.936	0.938	0.927
LIVE-II	PLCC	0.899	0.909	0.912	0.905	0.936	0.937
LIVE-II	SROCC	0.864	0.883	0.880	0.872	0.926	0.931
NBU-MDSID I	PLCC	0.960	0.950	0.952	0.959	0.955	0.963
NBU-MDSID I	SROCC	0.939	0.922	0.922	0.936	0.926	0.952
NBU-MDSID II	PLCC	0.807	0.702	0.729	0.810	0.748	0.859
NBU-MDSID II	SROCC	0.788	0.673	0.712	0.793	0.736	0.851

Table 5. Performance comparison among individual response and comprehensive response of five types of neurons

Database	Type	Neuron response
Database	Type	ODF TE	ODF TI	ODF ODD	RPC TE	RPC ODD
LIVE-I	JPEG	0.730	0.671	0.702	0.765	0.751
	JP2K	0.886	0.892	0.874	0.895	0.881
	WN	0.901	0.885	0.908	0.902	0.915
	GB	0.880	0.864	0.874	0.885	0.884
	FF	0.762	0.787	0.758	0.767	0.713
LIVE-II	JPEG	0.719	0.781	0.722	0.716	0.798
	JP2K	0.762	0.819	0.744	0.768	0.885
	WN	0.948	0.917	0.947	0.945	0.943
	GB	0.927	0.932	0.915	0.931	0.920
	FF	0.861	0.916	0.880	0.857	0.916

Table 6. SROCC results of different distortion types under five types of neuron responses

Algorithm	Method in Ref.[10]	Method in Ref. [3]	Method in Ref. [4]	Method in Ref. [13]	Method in Ref. [22]	Proposed method
Time /s	20.799	31.295	0.542	153.356	7.760	8.703

Table 7. Running time of each algorithm

Mengmeng Ye, Jinbin Hu, Xuejin Wang, Feng Shao. No-Reference Stereoscopic Image Quality Assessment Based on Binocular Neuron Response[J]. Laser & Optoelectronics Progress, 2021, 58(24): 2410007

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information