Foveated JND Model Based on Stereo Vision and Its Application in Image Compression with Hardware Implementation

Yuan Ji; Zhijie Zheng; hao Wu; Yin Zhang; Wendong Chen; Tingzhou Mu

doi:10.3788/AOS202141.1210001

Acta Optica Sinica, Volume. 41, Issue 12, 1210001(2021)

Foveated JND Model Based on Stereo Vision and Its Application in Image Compression with Hardware Implementation

Yuan Ji^1,2, Zhijie Zheng^1、*, hao Wu¹, Yin Zhang¹, Wendong Chen², and Tingzhou Mu²

Author Affiliations

¹Microelectronics Research and Development Center, Shanghai University, Shanghai 200072, China

²School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200072, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Test image of luminance masking characteristic experiment

Download full size

Fig. 2. Principles of 3D imaging

Download full size

Fig. 3. Test image of contrast masking characteristic experiment

Download full size

Fig. 4. Experimental results of luminance masking characteristic experiment. (a) Relationship between JND and luminance; (b) relationship between JND and eccentricity angle; (c) relationship between JND and depth; (d) relationship among JND, luminance, and eccentricity angle at d=1.0; (e) relationship among JND, luminance, and depth at e=20; (f) 4D image of luminance masking characteristic experiment

Download full size

Fig. 5. Experimental results of contrast masking characteristic experiment. (a) Relationship between JND and contrast; (b) relationship among JND, luminance, and contrast

Download full size

Fig. 6. Noise injection distributions of different JND models. (a) Original image; (b) proposed model; (c) Ref. [4]; (d) Ref. [3]

Download full size

Fig. 7. Enlarged view of central area of picture after pollution by different JND models. (a) Original image; (b) proposed model; (c) Ref. [4]; (d) Ref. [3]

Download full size

Fig. 8. Hardware system architecture framework based on FPGA

Download full size

Fig. 9. Hardware test platform

Download full size

Fig. 10. Comparison of image compression effect and local enlarged view. (a) Original images; (b) compressed images

Download full size

Table 1. Specific values of luminance masking characteristic and contrast masking characteristic experiments

View table

Table 1. Specific values of luminance masking characteristic and contrast masking characteristic experiments

Attribute	Luminance masking	Contrast masking
Background luminance	0,10,20,60,100,140,180,200,220,230,240	100,140,180,220
Contrast luminance	0	20,40,60,80
Eccentricity /(°)	1,2,3,5,10,15,20,25,30,35,40,45,50	1,2,3,5,10,15,20,25,30,35,40,45,50
Depth	0.2,0.4,0.6,0.8,1.0	0.2,0.4,0.6,0.8,1.0
Disparity /pixel	600,300,200,150,120	600,300,200,150,120

Table 2. Comparison of average subjective scores of different models

View table

Table 2. Comparison of average subjective scores of different models

Image	PSNR /dB	MOS
Image	PSNR /dB	Ref. [4]	Ref. [3]	Proposedmodel
Gate	32.01	4.20	3.50	4.80
Woman	32.33	4.30	3.50	4.90
Column	30.89	4.20	3.40	4.70
Hall	30.72	4.30	3.50	4.90
Lab	30.71	4.00	3.30	4.60
Repair	32.48	4.30	3.40	4.80
Report	30.24	4.20	3.30	4.80
Soccer	31.22	4.40	3.60	5.00
Branch	31.42	4.30	3.40	4.90
Umbrella	32.03	4.20	3.40	4.80
Average	31.41	4.24	3.43	4.82

Table 3. Color scale of different JNCD values
View table
Table 3. Color scale of different JNCD values
JNCD value Color scale Bit width
(0,4] 256 8
(4,6] 128 7
(6,8] 64 6
(8,10] 32 5
(10,11] 16 4
(11,∞] 8 3

Table 4. Evaluation index and bit rate of color image compression

View table

Table 4. Evaluation index and bit rate of color image compression

Image	PSNR /dB					FPSNR /dB	SSIM	Bitrate /%
Image	e is 0°--5°	e is 5°--10°	e is 10°--15°	e is 15°--30°	e>30°	FPSNR /dB	SSIM	Bitrate /%
Art	70.5936	58.9329	57.1360	52.1277	33.2020	40.4823	0.8675	61.59
Book	55.2470	51.6498	52.2067	49.4993	32.8974	39.8006	0.8686	57.15
Computer	65.8224	54.3767	53.7429	52.6272	33.1890	40.4265	0.8783	60.84
Doll	66.6142	54.6717	55.0177	51.9896	33.3387	40.6038	0.8799	61.53
Drumstick	62.9970	57.3913	56.8413	53.1031	33.7317	41.0152	0.8980	62.02
Dwarves	77.8636	62.7846	62.1044	53.2000	33.5468	40.9747	0.8669	63.92
Laundry	65.5591	53.5276	54.2809	52.7823	33.7454	41.0754	0.8861	62.26
Moebius	73.0924	66.3706	60.1905	53.2447	33.3951	40.6560	0.8637	61.78
Reindeer	64.3097	56.8130	58.1696	52.8886	33.3697	40.8316	0.8606	63.74
Average	66.8999	57.3909	56.6322	52.3847	33.3795	40.6518	0.8744	61.65

Tools

Get Citation

Copy Citation Text

Yuan Ji, Zhijie Zheng, hao Wu, Yin Zhang, Wendong Chen, Tingzhou Mu. Foveated JND Model Based on Stereo Vision and Its Application in Image Compression with Hardware Implementation[J]. Acta Optica Sinica, 2021, 41(12): 1210001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites