基于改進的自適應差分演化算法的二維Otsu多閾值圖像分割

羅鈞; 楊永松; 侍寶玉

doi:10.11999/JEIT180949

基于改進的自適應差分演化算法的二維Otsu多閾值圖像分割

doi: 10.11999/JEIT180949

重慶大學光電技術及系統(tǒng)教育部重點實驗室??重慶??400030

詳細信息

作者簡介:
羅鈞：男，1963年生，教授，博士生導師，研究方向為模式識別與人工智能，精密機械及測試計量，智能信息處理

楊永松：男，1994年生，碩士生，研究方向為嵌入式系統(tǒng)，機器視覺

侍寶玉：女，1994年生，碩士生，研究方向為嵌入式系統(tǒng)，機器視覺

通訊作者:
羅鈞　luojun@cqu.edu.cn

中圖分類號: TP391.41
計量
- 文章訪問數(shù): 3696
- HTML全文瀏覽量: 1392
- PDF下載量: 180
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2018-10-12
- 修回日期: 2019-03-04
- 網(wǎng)絡出版日期: 2019-03-28
- 刊出日期: 2019-08-01

Multi-threshold Image Segmentation of 2D Otsu Based on Improved Adaptive Differential Evolution Algorithm

Key Laboratory of Optoelectronic Technology and System of Ministry of Education, Chongqing University, Chongqing 400030, China

摘要

摘要: 針對常規(guī)最大類間方差法在多閾值圖像分割中存在的運算量大、計算時間長、分割精度較低等問題，該文提出一種基于改進的自適應差分演化(JADE)算法的2維Otsu多閾值分割法。首先，為增強初始化種群的質量、提升控制參數(shù)的適應性，將混沌映射機制融入到JADE算法中；進而，通過該改進算法求解2維 Otsu 多閾值圖像的最佳分割閾值；最終，將該算法與差分進化(DE), JADE，改進正弦參數(shù)自適應的差分進化(LSHADE-cnEpSin)以及增強的適應性微分變換差分進化(EFADE) 4種算法的2維Otsu多閾值圖像分割進行比較。實驗結果表明，與其它4種算法相比，基于改進JADE算法的2維Otsu多閾值圖像分割在分割速度以及精度上均有較明顯的改善。
- 圖像分割 /
- 最大類間方差法 /
- 混沌映射 /
- 改進的自適應差分演化算法
Abstract: The multi-threshold image segmentation of the classical 2D maximal between-cluster variance method has deficiencies such as large computation, long calculation time, low segmentation precision and so on. A multi-threshold segmentation of 2D Otsu based on improved Adaptive Differential Evolution (JADE) algorithm is proposed. Firstly, in order to enhance the quality of the initialized population and improve the adaptability of the control parameters, the chaotic mapping mechanism is integrated into the JADE algorithm. Furthermore, the optimal segmentation threshold of 2D Otsu multi-threshold image is solved by improved JADE algorithm. Finally, the algorithm is compared with multi-threshold image segmentation method of 2D Otsu based on Differential Evolution (DE), JADE, Improved Differential Evolution with Adaptive Sinusoidal Parameters (LSHADE-cnEpSin) and Enhanced Adaptive Differential Transformation Differential Evolution (EFADE) algorithm. The experimental results show that compared with the other four algorithms, the multi-threshold image segmentation of 2D Otsu based on the improved JADE algorithm has a significant improvement in terms of segmentation speed and accuracy.
- Image segmentation /
- Maximum interclass variance method /
- Chaotic map /
- Improved Adaptive Differential Evolution (JADE) algorithm

HTML全文

圖 1 2維多閾值分割直方圖

下載: 全尺寸圖片幻燈片

圖 2 基于CJADE算法2維Otsu多閾值分割方法流程圖

下載: 全尺寸圖片幻燈片

圖 3 分割效果圖

下載: 全尺寸圖片幻燈片

圖 4 進化曲線圖

下載: 全尺寸圖片幻燈片

表 1 算法1：混沌映射更新參數(shù)u_F和u_CR的偽代碼

　(1) If $\;\alpha < \beta $

　(2) 　　${u_{\rm CR}} = {u_1} \cdot {u_{\rm CR}} \cdot (1 - {u_{\rm CR}})$

　(3) 　　${u_F} = {u_2} \cdot {u_F} \cdot (1 - {u_F})$

　(4) Else

　(5) 　　${u_{\rm CR}} = (1 - c) \cdot {u_{\rm CR}} + c \cdot {{\rm mean}_{\rm A}}({S_{\rm CR}})$

　(6) 　　${u_F} = (1 - c) \cdot {u_F} + c \cdot {{\rm mean}_{\rm L}}({S_F})$

　(7) End If

下載: 導出CSV

表 2 PSNR、運算時間以及迭代次數(shù)的對比

算法		Lena (512$ \times $512)			Finger (256$ \times $256)			Pepper (512$ \times $512)
算法		2閾值	3閾值	4閾值	2閾值	3閾值	4閾值	2閾值	3閾值	4閾值
DE算法	PSNR(dB)	10.58	13.88	15.64	12.02	12.45	14.14	11.68	15.84	16.54
	收斂時間(s)	7.79	7.82	7.84	3.64	3.58	3.73	8.49	8.34	8.82
	迭代次數(shù)	72	58	64	62	57	66	45	43	47
JADE算法	PSNR(dB)	11.79	14.25	16.02	12.35	13.02	14.26	11.71	16.32	16.71
	收斂時間(s)	0.85	0.83	0.77	0.51	0.53	0.57	0.81	0.80	0.83
	迭代次數(shù)	52	54	50	59	62	58	60	56	58
LSHADE-cnEpSin算法	PSNR(dB)	13.70	14.98	15.67	12.07	12.77	14.46	12.23	16.19	17.02
	收斂時間(s)	0.79	0.75	0.82	0.45	0.48	0.46	0.78	0.82	0.78
	迭代次數(shù)	34	35	33	65	45	60	50	48	46
EFADE算法	PSNR(dB)	12.89	15.05	15.45	13.23	12.61	13.24	12.11	15.57	16.67
	收斂時間(s)	0.99	1.12	1.10	0.77	0.76	0.83	1.24	1.31	1.29
	迭代次數(shù)	45	42	46	50	48	52	40	38	41
CJADE算法	PSNR(dB)	13.93	15.64	16.25	13.65	14.67	14.89	12.56	16.57	17.12
	收斂時間(s)	0.64	0.66	0.65	0.45	0.44	0.48	0.61	0.64	0.66
	迭代次數(shù)	38	35	38	41	40	44	40	36	38

下載: 導出CSV

表 3 閾值和距離測度值的對比

算法		Lena (512$ \times $512)			Finger (256$ \times $256)			Pepper (512$ \times $512)
算法		2閾值	3閾值	4閾值	2閾值	3閾值	4閾值	2閾值	3閾值	4閾值
DE算法	距離測度	4645.67	4698.86	4747.74	1223.45	1247.75	1296.25	5340.87	5407.71	5513.28
DE算法	閾值	(68,71) (117,153)	(30, 32) (86,138) (193,199)	(88,95) (119,123) (151,153) (202,207)	(39,53) (155,165)	(108,124) (147,152) (168,180)	(23,38) (102,133) (150,157) (169,170)	(70,70) (117,161)	(84, 85) (142,162) (201,203)	(70,77) (111,112) (126,129) (129,179)
JADE算法	距離測度	4842.77	4912.21	4924.13	1315.43	1320.35	1326.23	5798.46	5822.86	5892.86
JADE算法	閾值	(89,149) (193,195)	(77,79) (114,149) (196,196)	(70,77) (109,137) (149,154) (182,183)	(138,166) (175,175)	(10,67) (143,164) (174,175)	(40,52) (50,110) (156,156) (171,172)	(88,91) (127,169)	(98, 115) (140,140) (178,178)	(96,101) (114,133) (149,150) (152,171)
LSHADE-cnEpSin算法	距離測度	4862.49	4905.97	4995.04	1256.65	1268.79	1289.32	5797.85	5899.34	5909.58
LSHADE-cnEpSin算法	閾值	(88,149) (194,195)	(79,79) (115,145) (177,177)	(76,76) (119,141) (158,160) (197,197)	(88,102) (183,183)	(64,82) (148,164) (183,184)	(36,39) (42,98) (145,155) (164,169)	(78,79) (126,177)	(84, 85) (127,159) (194,194)	(76,77) (121,122) (126,157) (192,193)
EFADE算法	距離測度	4848.87	4951.82	4973.23	1257.29	1267.42	1324.41	5788.61	5885.72	5892.13
EFADE算法	閾值	(89,148) (186,186)	(76,80) (130,153) (205,205)	(79,86) (112,137) (139,151) (201,203)	(44,51) (142,176)	(30,42) (140,162) (172,174)	(41,46) (68,83) (141,167) (172,173)	(86,90) (120,176)	(73, 74) (121,159) (193,194)	(53,55) (121,123) (154,154) (180,182)
CJADE算法	距離測度	4863.53	4977.34	4999.63	1327.84	1329.17	1331.28	5799.13	5898.73	5912.18
CJADE算法	閾值	(87,149) (194,194)	(77,78) (115,148) (194,195)	(78,80) (117,139) (156,156) (199,199)	(143,166) (173,173)	(25, 62) (142,166) (174,175)	(40,45) (70,98) (156,158) (162,162)	(84,85) (124,173)	(77, 78) (123,164) (195,195)	(54,55) (99,100) (129,160) (199,199)

下載: 導出CSV

參考文獻(17)

劉健莊, 栗文青. 灰度圖象的二維Otsu自動閾值分割法[J]. 自動化學報, 1993, 19(1): 101–105. doi: 10.16383/j.aas.1993.01.015

LIU Jianzhuang and LI Wenqing. The automatic thresholding of gray-level pictures via two-dimensional otsu method[J]. Acta Automatica Sinica, 1993, 19(1): 101–105. doi: 10.16383/j.aas.1993.01.015

申鉉京, 劉翔, 陳海鵬. 基于多閾值Otsu準則的閾值分割快速計算[J]. 電子與信息學報, 2017, 39(1): 144–149. doi: 10.11999/JEIT160248

SHEN Xuanjing, LIU Xiang, and CHEN Haipeng. Fast computation of threshold based on multi-threshold Otsu criterion[J]. Journal of Electronics &Information Technology, 2017, 39(1): 144–149. doi: 10.11999/JEIT160248

HU Min, LI Mei, and WANG Ronggui. Application of an improved Otsu algorithm in image segmentation[J]. Journal of Electronic Measurement and Instrument, 2010, 24(5): 443–449. doi: 10.3724/SP.J.1187.2010.00443

ZHANG Jingqiao and SANDERSON A C. JADE: Adaptive differential evolution with optional external archive[J]. IEEE Transactions on Evolutionary Computation, 2009, 13(5): 945–958. doi: 10.1109/TEVC.2009.2014613

TANABE R and FUKUNAGA A. Success-history based parameter adaptation for differential evolution[C]. 2013 IEEE Congress on Evolutionary Computation, Cancun, Mexico, 2013: 71–78.

TANABE R and FUKUNAGA A S. Improving the search performance of SHADE using linear population size reduction[C]. 2014 IEEE Congress on Evolutionary Computation, Beijing, China, 2014: 1658–1665.

AWAD N H, ALI M Z, SUGANTHAN P N, et al. An ensemble sinusoidal parameter adaptation incorporated with L-SHADE for solving CEC2014 benchmark problems[C]. 2016 IEEE Congress on Evolutionary Computation, Vancouver, Canada, 2016: 2958–2965.

AWAD N H, ALI M Z, and SUGANTHAN P N. Ensemble sinusoidal differential covariance matrix adaptation with Euclidean neighborhood for solving CEC2017 benchmark problems[C]. 2017 IEEE Congress on Evolutionary Computation, San Sebastian, Spain, 2017: 372–379.

MOHAMED A W and SUGANTHAN P N. Real-parameter unconstrained optimization based on enhanced fitness-adaptive differential evolution algorithm with novel mutation[J]. Soft Computing, 2018, 22(10): 3215–3235. doi: 10.1007/s00500-017-2777-2

STHITPATTANAPONGSA P and SRINARK T. A two-stage Otsu’S thresholding based method on a 2D histogram[C]. IEEE 7th International Conference on Intelligent Computer Communication and Processing, Cluj-Napoca, Romania, 2011: 345–348.

張春美, 陳杰, 辛斌. 參數(shù)適應性分布式差分進化算法[J]. 控制與決策, 2014, 29(4): 701–706. doi: 10.13195/j.kzyjc.2013.0080

ZHANG Chunmei, CHEN Jie, and XIN Bin. Distributed differential evolution algorithm with adaptive parameters[J]. Control and Decision, 2014, 29(4): 701–706. doi: 10.13195/j.kzyjc.2013.0080

王李進, 鐘一文, 尹義龍. 帶外部存檔的正交交叉布谷鳥搜索算法[J]. 計算機研究與發(fā)展, 2015, 52(11): 2496–2507. doi: 10.7544/issn1000-1239.2015.20148042

WANG Lijin, ZHONG Yiwen, and YIN Yilong. Orthogonal crossover cuckoo search algorithm with external archive[J]. Journal of Computer Research and Development, 2015, 52(11): 2496–2507. doi: 10.7544/issn1000-1239.2015.20148042

RERE L M R, FANANY M I, and MURNI A. Adaptive DE based on chaotic sequences and random adjustment for image contrast enhancement[C]. 2014 International Conference of Advanced Informatics: Concept, Theory and Application, Bandung, Indonesia, 2015: 220–225. doi: 10.1109/ICAICTA.2014.7005944.

陳志剛, 梁滌青, 鄧小鴻, 等. Logistic混沌映射性能分析與改進[J]. 電子與信息學報, 2016, 38(6): 1547–1551. doi: 10.11999/JEIT151039

CHEN Zhigang, LIANG Diqing, DENG Xiaohong, et al. Performance analysis and improvement of logistic chaotic mapping[J]. Journal of Electronics &Information Technology, 2016, 38(6): 1547–1551. doi: 10.11999/JEIT151039

陳如清. 采用新型粒子群算法的電力電子裝置在線故障診斷方法[J]. 中國電機工程學報, 2008, 28(24): 70–74. doi: 10.3321/j.issn:0258-8013.2008.24.012

CHEN Ruqing. A novel PSO based on-line fault diagnosis approach for power electronic system[J]. Proceedings of the CSEE, 2008, 28(24): 70–74. doi: 10.3321/j.issn:0258-8013.2008.24.012

SHA Chunshi, HOU Jian, and CUI Hongxia. A robust 2D Otsu’s thresholding method in image segmentation[J]. Journal of Visual Communication and Image Representation, 2016, 41: 339–351. doi: 10.1016/j.jvcir.2016.10.013

HUYNH-THU Q and GHANBARI M. The accuracy of PSNR in predicting video quality for different video scenes and frame rates[J]. Telecommunication Systems, 2012, 49(1): 35–48. doi: 10.1007/s11235-010-9351-x

施引文獻

資源附件(0)

訪問統(tǒng)計