Robust lockwire segmentation with multiscale boundary-driven regional stability

Yanxia Xie; Junhua Sun

doi:10.1364/JOSAA.472215

Journal of the Optical Society of America A
Vol. 40,
Issue 3,
pp. 397-410
(2023)
•https://doi.org/10.1364/JOSAA.472215

Robust lockwire segmentation with multiscale boundary-driven regional stability

Yanxia Xie and Junhua Sun

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Yanxia Xie and Junhua Sun, "Robust lockwire segmentation with multiscale boundary-driven regional stability," J. Opt. Soc. Am. A 40, 397-410 (2023)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Abstract

Lockwire segmentation plays a vital role in ensuring mechanical safety in industrial fields. Aiming at the missed detection problem encountered in blurred and low-contrast situations, we propose a robust lockwire segmentation method based on multiscale boundary-driven regional stability. We first design a novel multiscale boundary-driven stability criterion to generate a blur-robustness stability map. Then, the curvilinear structure enhancement metric and linearity measurement function are defined to compute the likeliness of stable regions to belong to lockwires. Finally, the closed boundaries of lockwires are determined to achieve accurate segmentation. Experimental results demonstrate that our proposed method outperforms state-of-the-art object segmentation methods.

Full Article | PDF Article

More Like This

Object-of-interest image segmentation based on human attention and semantic region clustering

Byoung Chul Ko and Jae-Yeal Nam
J. Opt. Soc. Am. A 23(10) 2462-2470 (2006)

Fusion network based on the dual attention mechanism and atrous spatial pyramid pooling for automatic segmentation in retinal vessel images

Bingtao Liang, Chen Tang, Min Xu, Tianbo Wu, and Zhenkun Lei
J. Opt. Soc. Am. A 39(8) 1393-1402 (2022)

Surface weak scratch detection for optical elements based on a multimodal imaging system and a deep encoder–decoder network

Xiao Liang, Jingshuang Sun, Xuewei Wang, Jie Li, Lianpeng Zhang, and Jingbo Guo
J. Opt. Soc. Am. A 40(6) 1237-1248 (2023)

Data availability

Data underlying the region detector test results presented in this paper are available in Ref. [36]. Data underlying the lockwire segmentation results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.

36. K. Mikolajczyk, T. Tuytelaars, C. Schmid, A. Zisserman, J. Matas, F. Schaffalitzky, T. Kadir, and L. Van Gool, “A comparison of affine region detectors,” Int. J. Comput. Vis. 65, 43–72 (2005). [CrossRef]

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (11)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (6)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (20)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Method	Parameter Settings
Ours	Number of orientations $K \in {4, 5, 6, 7, 8}$ , scale $σ_{i} \in {1.5, 2, 3, 4}$ , area threshold $ε_{a} = 5$ , perimeter threshold $ε_{l} = 2$ , density threshold $d = 1000$ .
Graph Cut [11]	Input image is scaled down to 25%
GrabCut [10]	Input image is scaled down to 25%
Level set [15]	Weight of length term $β = 10$ ; weight of distance term $μ = 0.1$ ; parameter $α = 0.1$
MSER-based [37]	Min area = 5; max area = 200, stability range $δ \in {5, 6, 7, 8, 9}$
U-Net [17]	Batch size = 4; initial learning rate = 0.001; learning rate decay = 0.3; epoch = 100
$U^{2}$ -Net [19]	Batch size = 12; learning rate = 0.001; epoch = 1000
CaraNet [28]	Batch size = 6; initial learning rate = 0.001; learning rate decay = 0.1; epoch = 300.

	Images <1, 2>		Images <1, 3>		Images <1, 4>		Images <1, 5>		Images <1, 6>
Scene	MSER	Ours	MSER	Ours	MSER	Ours	MSER	Ours	MSER	Ours
Bark	0.26	0.44	0.16	0.42	0.15	0.52	0.03	0.45	0.00	0.34
Boat	0.65	0.79	0.58	0.74	0.42	0.52	0.38	0.45	0.12	0.20
Bikes	0.77	0.86	0.77	0.83	0.59	0.76	0.49	0.68	0.38	0.59
Trees	0.39	0.50	0.33	0.45	0.22	0.28	0.17	0.20	0.08	0.12
Bricks	0.63	0.74	0.56	0.60	0.47	0.53	0.36	0.44	0.18	0.28
Graffiti	0.67	0.78	0.58	0.69	0.56	0.67	0.46	0.60	0.36	0.38
Cars	0.86	0.92	0.83	0.88	0.73	0.78	0.56	0.67	0.19	0.45
Ubc	0.84	0.92	0.80	0.90	0.73	0.85	0.60	0.79	0.43	0.61

	Images <1, 2>		Images <1, 3>		Images <1, 4>		Images <1, 5>		Images <1, 6>
Scene	MSER	Ours	MSER	Ours	MSER	Ours	MSER	Ours	MSER	Ours
Bark	0.41	0.28	0.34	0.31	0.31	0.47	0.23	0.40	0.14	0.32
Boat	0.56	0.68	0.54	0.66	0.47	0.50	0.47	0.47	0.29	0.30
Bikes	0.73	0.77	0.76	0.75	0.68	0.71	0.66	0.65	0.63	0.57
Trees	0.40	0.44	0.36	0.38	0.28	0.26	0.23	0.19	0.17	0.13
Bricks	0.49	0.64	0.47	0.53	0.43	0.46	0.39	0.40	0.33	0.33
Graffiti	0.65	0.71	0.59	0.62	0.58	0.61	0.52	0.53	0.44	0.34
Cars	0.78	0.89	0.77	0.84	0.70	0.74	0.64	0.58	0.36	0.36
ubc	0.45	0.60	0.58	0.73	0.69	0.78	0.75	0.84	0.78	0.88

		Images <1, 2>		Images <1, 3>		Images <1, 4>
Scene	Gaussian Noise $σ^{2}$	MSER	Ours	MSER	Ours	MSER	Ours
Bark	0.01	0.13	0.29	0.07	0.18	0.04	0.18
	0.03	0.06	0.17	0.01	0.08	0.01	0.08
	0.05	0.03	0.08	0.01	0.04	0.01	0.04
Trees	0.01	0.29	0.48	0.25	0.43	0.17	0.33
	0.03	0.19	0.43	0.16	0.36	0.13	0.30
	0.05	0.14	0.40	0.12	0.32	0.09	0.27
Bricks	0.01	0.40	0.60	0.35	0.48	0.28	0.42
	0.03	0.21	0.35	0.17	0.32	0.15	0.23
	0.05	0.13	0.24	0.09	0.21	0.06	0.16
Graffiti	0.01	0.53	0.68	0.53	0.63	0.42	0.66
	0.03	0.43	0.65	0.38	0.56	0.35	0.56
	0.05	0.33	0.59	0.28	0.53	0.20	0.52
ubc	0.01	0.74	0.90	0.69	0.90	0.70	0.90
	0.03	0.48	0.89	0.47	0.88	0.45	0.84
	0.05	0.29	0.84	0.30	0.80	0.26	0.81

		Images <1, 2>		Images <1, 3>		Images <1, 4>
Scene	Gaussian Noise $σ^{2}$	MSER	Ours	MSER	Ours	MSER	Ours
Bark	0.01	0.28	0.26	0.22	0.19	0.17	0.19
	0.03	0.18	0.21	0.10	0.14	0.08	0.14
	0.05	0.14	0.17	0.09	0.11	0.06	0.11
Trees	0.01	0.33	0.44	0.30	0.40	0.23	0.33
	0.03	0.25	0.42	0.22	0.37	0.20	0.33
	0.05	0.20	0.42	0.19	0.35	0.17	0.32
Bricks	0.01	0.37	0.52	0.36	0.43	0.32	0.40
	0.03	0.27	0.33	0.27	0.32	0.22	0.31
	0.05	0.20	0.26	0.20	0.29	0.15	0.26
Graffiti	0.01	0.57	0.65	0.58	0.60	0.50	0.61
	0.03	0.48	0.63	0.45	0.55	0.44	0.55
	0.05	0.41	0.58	0.37	0.53	0.30	0.54
ubc	0.01	0.74	0.86	0.70	0.86	0.71	0.86
	0.03	0.53	0.85	0.52	0.85	0.50	0.79
	0.05	0.38	0.81	0.39	0.77	0.34	0.78

Types	Method	mIoU
Classic segmentation	Ours	$0.73 \pm 0.01$
methods	Graph Cut [11]	$0.46 \pm 0.00$
	GrabCut [10]	$0.49 \pm 0.00$
	level set [15]	$0.48 \pm 0.01$
	MSER-based method [37]	$0.67 \pm 0.02$
CNN-based methods	U-Net [17]	$0.67 \pm 0.03$
	$U^{2}$ -Net [19]	$0.67 \pm 0.02$
	CaraNet [28]	$0.63 \pm 0.01$

Abstract

Data availability

Cited By

Figures (11)

Tables (6)

Equations (20)

Journal of the Optical Society of America A