Image Segmentation of Non-Destructive Test Based on Image Patch and Cluster Information Quantity

Zhanlong Zhu; Yongjun Liu; Yamei Li; Junfen Wang; Boyuan Deng

doi:10.3788/LOP202158.1210009

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 12 >Page 1210009 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 12, 1210009 (2021)

Image Segmentation of Non-Destructive Test Based on Image Patch and Cluster Information Quantity

Zhanlong Zhu^1、2、3, Yongjun Liu¹, Yamei Li^1、2, Junfen Wang^1、2、*, and Boyuan Deng¹

Author Affiliations

¹School of Information Engineering, Heibei GEO University, Shijiazhuang, Hebei 0 50031, China

²Hebei Key Laboratory of Optoelectronic Information and Geo-Detection Technology, Shijiazhuang, Hebei 0 50031, China

³Intelligent Sensor Network Engineering Research Center of Hebei Province, Shijiazhuang, Hebei 0 50031, China

show less

DOI: 10.3788/LOP202158.1210009 Cite this Article Set citation alerts

Zhanlong Zhu, Yongjun Liu, Yamei Li, Junfen Wang, Boyuan Deng. Image Segmentation of Non-Destructive Test Based on Image Patch and Cluster Information Quantity[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210009 Copy Citation Text

show less

Examples of pixel weight setting in image patch. (a) Three image patches called A, B, and C, respectively; (b) gray values of image patch A; (c) gray values of image patch B; (d) gray values of image patch C; (e) ωjr in image patch A; (f) ωjr in image patch B; (g) ωjr in image patch C

Fig. 1. Examples of pixel weight setting in image patch. (a) Three image patches called A, B, and C, respectively; (b) gray values of image patch A; (c) gray values of image patch B; (d) gray values of image patch C; (e) ω_jr in image patch A; (f) ω_jr in image patch B; (g) ω_jr in image patch C

Download full size

NDT images and their gray histograms. (a) Image of #NDT1; (b) image of #NDT2; (c) image of #NDT3; (d) image of #NDT4; (e) image of #NDT5; (f) image of #NDT6; (g) gray histogram of #NDT1; (h) gray histogram of #NDT2; (i) gray histogram of #NDT3; (j) gray histogram of #NDT4; (k) gray histogram of #NDT5; (l) gray histogram of #NDT6

Fig. 2. NDT images and their gray histograms. (a) Image of #NDT1; (b) image of #NDT2; (c) image of #NDT3; (d) image of #NDT4; (e) image of #NDT5; (f) image of #NDT6; (g) gray histogram of #NDT1; (h) gray histogram of #NDT2; (i) gray histogram of #NDT3; (j) gray histogram of #NDT4; (k) gray histogram of #NDT5; (l) gray histogram of #NDT6

Download full size

Fig. 3. Segmentation results of #NDT1 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Fig. 4. Segmentation results of #NDT2 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Fig. 5. Segmentation results of #NDT3 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Fig. 6. Segmentation results of #NDT4 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Fig. 7. Segmentation results of #NDT5 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Fig. 8. Segmentation results of #NDT6 for GN(0,0.01). (a) Noisy image; (b) standard segmentation image; (c) result of NDFCM algorithm; (d) result of FCM_S1 algorithm;(e) result of FCM_S2 algorithm; (f) result of WIPFCM algorithm; (g) result of KCWFLICM algorithm; (h) result of IS-FCM algorithm; (i) result of method in Ref.[14]; (j) result of ICPFCM algorithm

Download full size

Algorithm	Parameter
Algorithm	m	α	β	Z	λ_α	q	λ_s	λ_g	ε	t
NDFCM^[2]	2	∥	∥	200	1	3	3	3	10^-4	∥
FCM_S1^[6]	2	∥	6	200	∥	3	∥	∥	10^-4	∥
FCM_S2^[6]	2	∥	6	200	∥	3	∥	∥	10^-4	∥
WIPFCM^[9]	2	∥	∥	200	∥	3	∥	∥	10^-4	∥
KCWFLICM^[10]	2	∥	∥	200	∥	3	∥	∥	10^-4	∥
IS-FCM^[11]	2	0.5	∥	200	∥	∥	∥	∥	10^-4	∥
Algorithm in Ref.[14]	2	∥	∥	200	∥	∥	∥	∥	10^-4	∥
ICPFCM	2	∥	∥	200	∥	3	∥	∥	10^-4	10

Table 1. Parameter setting of correlated algorithms

Algorithm	Noise level	#NDT1		#NDT2		#NDT3		#NDT4		#NDT5		#NDT6
Algorithm	Noise level	SA	ARI	SA	ARI	SA	ARI	SA	ARI	SA	ARI	SA	ARI
NDFCM	GN(0,0.01)	93.8	87.7	57.7	15.4	77.6	55.3	74.0	47.9	89.3	78.6	86.2	72.4
	GN(0,0.02)	91.5	83.0	52.9	5.7	74.3	48.6	66.6	33.2	87.5	75.0	84.8	69.7
	SPN(0.1)	94.1	88.1	57.4	14.8	78.9	57.8	98.4	96.9	89.0	80.0	91.1	82.2
	SPN(0.2)	90.7	81.4	54.9	9.7	75.5	50.9	67.2	34.4	88.0	76.0	81.3	62.6
FCM_S1	GN(0,0.01)	91.9	83.8	56.1	12.2	76.4	52.8	61.6	23.1	87.6	75.2	83.4	66.8
	GN(0,0.02)	86.1	72.3	53.7	7.4	72.8	45.6	60.0	19.9	84.5	69.1	80.8	61.6
	SPN(0.1)	83.8	67.5	59.2	18.4	71.8	43.7	60.9	21.7	80.4	60.7	79.9	59.7
	SPN(0.2)	73.2	46.3	57.8	15.5	64.1	28.1	60.5	21.0	76.6	53.1	68.4	36.9
FCM_S2	GN(0,0.01)	90.5	80.9	56.8	13.6	76.3	52.5	66.0	32.1	87.9	75.7	81.2	62.3
	GN(0,0.02)	84.1	68.1	53.6	7.1	71.2	42.4	61.9	23.9	85.3	70.5	80.2	60.4
	SPN(0.1)	92.8	85.6	51.4	2.9	82.3	64.7	58.0	16.0	91.5	83.1	93.6	87.1
	SPN(0.2)	89.7	79.4	49.9	-0.1	77.5	55.0	56.7	13.4	89.7	79.5	87.0	73.9
WIPFCM	GN(0,0.01)	84.2	68.4	54.0	8.0	75.8	51.6	59.9	19.8	86.0	72.0	78.5	57.0
	GN(0,0.02)	75.6	51.2	52.8	5.6	65.6	31.2	51.7	3.4	80.1	60.1	76.2	52.3
	SPN(0.1)	59.5	18.9	60.6	21.3	59.4	18.7	62.7	25.3	68.6	37.2	62.6	25.1
	SPN(0.2)	41.0	-18.0	36.4	-27.2	43.7	-12.5	44.1	-11.9	53.4	6.8	4.6	-10.8
KCWFLICM	GN(0,0.01)	95.3	90.7	68.6	37.1	81.1	62.3	98.7	97.5	90.8	81.8	90.7	81.3
	GN(0,0.02)	95.1	90.2	59.3	18.7	75.0	50.0	98.5	96.9	89.9	79.8	88.0	76.1
	SPN(0.1)	94.4	88.8	66.6	33.2	76.8	53.5	98.4	96.8	87.3	74.6	87.1	74.3
	SPN(0.2)	92.0	84.0	57.5	15.0	72.5	45.0	71.3	42.6	88.8	77.6	75.4	50.7
IS-FCM	GN(0,0.01)	93.5	86.9	75.8	51.6	89.3	78.5	95.9	91.9	84.1	68.3	90.3	80.6
	GN(0,0.02)	90.6	81.2	66.6	33.1	77.6	55.2	89.2	78.3	76.4	52.7	85.2	70.4
	SPN(0.1)	85.1	70.3	93.0	86.0	85.9	71.7	89.9	79.9	91.5	83.0	82.8	65.6
	SPN(0.2)	81.2	62.4	88.5	77.0	82.5	65.1	87.7	75.5	86.9	73.8	79.6	59.2
Algorithm	Noise level	#NDT1		#NDT2		#NDT3		#NDT4		#NDT5		#NDT6
Algorithm	Noise level	SA	ARI	SA	ARI	SA	ARI	SA	ARI	SA	ARI	SA	ARI
Method in Ref.[14]	GN(0,0.01)	91.7	83.3	61.8	23.6	77.9	55.9	75.5	51.0	82.0	64.1	78.3	56.5
	GN(0,0.02)	78.0	56.0	58.3	16.6	65.4	30.7	64.9	27.7	74.8	49.6	71.7	43.4
	SPN(0.1)	85.2	70.3	93.0	86.0	98.1	96.2	90.1	80.1	90.2	80.4	95.7	91.4
	SPN(0.2)	82.0	64.0	88.5	77.1	82.5	65.0	87.9	75.8	84.9	69.8	79.5	58.9
ICPFCM	GN(0,0.01)	98.7	97.3	98.7	97.4	90.6	81.3	99.0	98.0	94.2	88.4	98.2	96.3
	GN(0,0.02)	98.4	96.8	98.3	96.5	84.3	68.6	98.9	97.7	91.8	83.7	96.7	93.4
	SPN(0.1)	99.1	98.3	98.2	96.5	93.8	87.7	99.2	98.5	95.2	90.3	99.2	98.4
	SPN(0.2)	99.1	98.2	98.2	96.4	93.6	87.2	99.2	98.3	94.9	89.9	99.1	98.2

Table 2. SA and ARI values obtained by segmentation for images #NDT1--#NDT6%

Image (size of image)	NDFCM	FCM_S1	FCM_S2	WIPFCM	KCWFLICM	IS-FCM	Method in Ref. [14]	ICPFCM
#NDT1(131×232)	1.36	0.29	0.36	5.74	15.30	0.52	0.56	1.61
#NDT2 (60×166)	0.58	0.12	0.17	1.65	10.21	0.18	0.17	1.61
#NDT3(51×88)	0.32	0.11	0.17	1.18	9.18	0.11	0.16	0.82
#NDT4(56×271)	1.96	0.22	0.28	2.62	11.27	0.54	0.63	2.26
#NDT5(70×100)	0.36	0.11	0.16	1.31	8.91	0.15	0.12	1.20
#NDT6 (51×98)	0.33	0.10	0.16	1.23	8.73	0.15	0.17	1.01

Table 3. Running time of different algorithmss

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information