Video Denoising and Object Detection Based on RPCA Model with l1-TV Regularization Constraints

Guoliang Yang; Dingling Yu; Zhendong Lai

doi:10.3788/LOP57.161506

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 16 >Page 161506 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 16, 161506 (2020)

Video Denoising and Object Detection Based on RPCA Model with l₁-TV Regularization Constraints

Guoliang Yang, Dingling Yu^*, and Zhendong Lai

Author Affiliations

School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China

show less

DOI: 10.3788/LOP57.161506 Cite this Article Set citation alerts

Guoliang Yang, Dingling Yu, Zhendong Lai. Video Denoising and Object Detection Based on RPCA Model with l₁-TV Regularization Constraints[J]. Laser & Optoelectronics Progress, 2020, 57(16): 161506 Copy Citation Text

show less

Fig. 1. Framework of proposed method

Download full size

Gaussian noise detection results. (a)(b) Static background; (c)(d) dynamic background sequence; (e) turbulent environment sequence; (f) camera shake sequence; (g) high frame rate sequence

Fig. 2. Gaussian noise detection results. (a)(b) Static background; (c)(d) dynamic background sequence; (e) turbulent environment sequence; (f) camera shake sequence; (g) high frame rate sequence

Download full size

Fig. 3. Gaussian noise detection results with different variances.(a) 1969 frame and (b) 2000 frame in dynamic background; (c) 1689 frame and (d) 1969 frame in static background

Download full size

Algorithm 1:Video denoising and object detection based on RPCA model with l₁-TV regularization constraints

Input: Original Video V∈R^m^×ⁿ^×^p, then add noise to the video to form a noisy video. Stack the noisy video frames to form O∈R^mn^×^p;Initialization: L⁰, S⁰, E⁰, F⁰, X⁰, Y⁰=0∈R^mn^×^p, k=0, μ_max=1.25/‖Y‖_F, λ₁,λ₂,λ₃>0,μ⁰=10^-2,ρ=1.5;Output: Optimal solution(F^k, K^k)1. While not converged do2. Update L^k⁺¹ via Eq. (12); Update S^k⁺¹ via Eq. (15); Update E^k⁺¹ via Eq. (17);3. Update F^k⁺¹ via Eq. (23); Update X^k⁺¹ via Eq. (24); Update Y^k⁺¹ via Eq. (25);4. Update μ^k⁺¹=min(ρμ^k,μ_max);5. K=K+1;6. Check the convergence condition:

‖ O - L^{k} - F^{k} ‖_{F}^{2}

≤10^-8

‖ O ‖_{F}^{2}

;7. End

Table 1. Flow of proposed algorithm

Image	DECOLOR			TVRPCA			IBT			LR l₁ TV			Proposed
Image	P	R	F	P	R	F	P	R	F	P	R	F	P	R	F
Fig. 2(a)	0.32	0.54	0.44	0.83	0.57	0.69	0.65	0.51	0.57	0.46	0.70	0.63	0.73	0.89	0.80
Fig. 2(b)	0.84	0.45	0.59	0.79	0.45	0.57	0.50	0.63	0.56	0.72	0.80	0.76	0.88	0.72	0.79
Fig. 2(c)	0.90	0.51	0.65	0.90	0.83	0.86	0.63	0.65	0.64	0.46	0.60	0.51	0.90	0.86	0.88
Fig. 2(d)	0.70	0.69	0.69	0.82	0.51	0.66	0.62	0.31	0.54	0.73	0.87	0.80	0.70	0.93	0.85
Fig. 2(e)	0.22	0.65	0.32	0.78	0.71	0.75	0.72	0.80	0.76	0.73	0.86	0.79	0.83	0.87	0.85
Fig. 2(f)	0.85	0.53	0.68	0.82	0.78	0.80	0.53	0.27	0.36	0.75	0.85	0.80	0.89	0.94	0.91
Fig.2(g)	0.89	0.51	0.64	0.90	0.43	0.58	0.69	0.65	0.67	0.63	0.66	0.64	0.66	0.66	0.66

Table 2. Measurement values of three indicators of different algorithms

Method	Fig. 2(a)	Fig. 2(b)	Fig. 2(c)	Fig. 2(d)	Fig. 2(e)	Fig. 2(f)	Fig. 2(g)
Proposed	20.04	24.31	29.95	15.30	18.23	13.67	18.19
DECOLOR	30.83	29.12	37.90	25.98	17.98	23.95	37.63
IBT	57.44	48.64	92.56	87.63	76.54	89.63	77.98
TVRPCA	55.23	43.56	63.72	78.03	63.83	74.38	76.93
LRl₁TV	31.32	21.42	30.43	24.78	10.51	20.66	21.58

Table 3. Comparison of running time of five algorithmss

Guoliang Yang, Dingling Yu, Zhendong Lai. Video Denoising and Object Detection Based on RPCA Model with l₁-TV Regularization Constraints[J]. Laser & Optoelectronics Progress, 2020, 57(16): 161506

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information