田思洋等：局部二值模式在连铸坯表面缺陷识别中的应用 ·1733· 表面缺陷检测.智能系统学报，2012,7(1)：75) 3结论 B]Ojala T,Pietikainen M,Harwood D.A comparative study of tex- (1)LBP等价模式具有良好的稳定性和高效性， ture measures with classification based on featured distributions 对连铸坯常见表面缺陷的识别率可达到94.9%，但仍 Pattern Recognit,1996,29(1):51 4]Ojala T,Pietikainen M,Maenpia T.Multiresolution gray-scale 有提升的空间. and rotation invariant texture classification with local binary pat- (2)改进的MB-LBP,算法具有特征维度低、计算 terns.IEEE Trans Pattern Anal Mach Intell,2002,24(7):971 速度快、鲁棒性好、抗噪能力强以及缺陷识别准确率高 [5]Dunn D.Higgins W E.Optimal Gabor filters for texture segmenta- 的优点. tion.IEEE Trans Image Process,1995,4(7):947 (3)将改进的MB-LBP,算法应用于连铸坯表面 [6]Song K C,Yan Y H.A noise robust method based on completed 缺陷识别，对裂纹、划伤、压痕和凹坑等四类缺陷样本 local binary patters for hot-olled steel strip surface defects.Appl Surf Sci,2013,285:858 和无缺陷样本的分类准确率达到近95%，比传统LBP 7]Aghdam S R.Amid E,Imani M F.A fast method of steel surface 算法提高近6%，可以满足实际检测的需求. defect detection using decision trees applied to LBP based features// 2012 7th IEEE Conference on Industrial Electronics and Applica- 参考文献 tions (/CIEA).Piscataway:IEEE,2012:1447 8] Mansano M,Pavesi L,Oliveira LS,et al.Inspection of metallic [Xu K.Yang C L,Zhou P,et al.Online detection technique of surfaces using local binary patterns /IECON 201137th Annual surface cracks for continuous casting slabs based on linear lasers. Conference on IEEE Industrial Electronics Society.Piscataway: JUniv Sci Technol Beijing,2009,31(12)1620 EEE,2011:2227 (徐科，杨朝霖，周鹏，等.基于线型激光的连铸板坯表面裂 Pietikainen M,Ojala T,Xu Z.Rotation-invariant texture classifi- 纹在线检测技术.北京科技大学学报，2009,31(12)：1620) cation using feature distributions.Pattern Recognit,2000,33 Wu J W,Yan JQ,Fang Z H,et al.Defect detection on a steel (1):43 slab surface based on the characteristics of an image's saliency re- [10]Zhang L,Chu R F,Xiang S M,et al.Face Detection Based on gion.CAAI Trans Intell Syst,2012,7(1):75 Multi-block IBP Representation.Springer Berlin Heidelberg, (吴家伟，严京旗，方志宏，等.基于图像显著性特征的铸坯 2007:11田思洋等: 局部二值模式在连铸坯表面缺陷识别中的应用 3 结论 ( 1) LBP 等价模式具有良好的稳定性和高效性， 对连铸坯常见表面缺陷的识别率可达到 94. 9% ，但仍 有提升的空间． ( 2) 改进的 MB-LBPu2 8，1算法具有特征维度低、计算 速度快、鲁棒性好、抗噪能力强以及缺陷识别准确率高 的优点． ( 3) 将改进的 MB-LBPu2 8，1算法应用于连铸坯表面 缺陷识别，对裂纹、划伤、压痕和凹坑等四类缺陷样本 和无缺陷样本的分类准确率达到近 95% ，比传统 LBP 算法提高近 6% ，可以满足实际检测的需求． 参 考 文 献 ［1］ Xu K，Yang C L，Zhou P，et al． Online detection technique of surface cracks for continuous casting slabs based on linear lasers． J Univ Sci Technol Beijing，2009，31( 12) : 1620 ( 徐科，杨朝霖，周鹏，等． 基于线型激光的连铸板坯表面裂 纹在线检测技术． 北京科技大学学报，2009，31( 12) : 1620) ［2］ Wu J W，Yan J Q，Fang Z H，et al． Defect detection on a steel slab surface based on the characteristics of an image's saliency re￾gion． CAAI Trans Intell Syst，2012，7( 1) : 75 ( 吴家伟，严京旗，方志宏，等． 基于图像显著性特征的铸坯 表面缺陷检测． 智能系统学报，2012，7( 1) : 75) ［3］ Ojala T，Pietikinen M，Harwood D． A comparative study of tex￾ture measures with classification based on featured distributions． Pattern Ｒecognit，1996，29( 1) : 51 ［4］ Ojala T，Pietikinen M，Menp T． Multiresolution gray-scale and rotation invariant texture classification with local binary pat￾terns． IEEE Trans Pattern Anal Mach Intell，2002，24( 7) : 971 ［5］ Dunn D，Higgins W E． Optimal Gabor filters for texture segmenta￾tion． IEEE Trans Image Process，1995，4( 7) : 947 ［6］ Song K C，Yan Y H． A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects． Appl Surf Sci，2013，285: 858 ［7］ Aghdam S Ｒ，Amid E，Imani M F． A fast method of steel surface defect detection using decision trees applied to LBP based features / / 2012 7th IEEE Conference on Industrial Electronics and Applica￾tions ( ICIEA) ． Piscataway: IEEE，2012: 1447 ［8］ Mansano M，Pavesi L，Oliveira L S，et al． Inspection of metallic surfaces using local binary patterns / / IECON 2011-37th Annual Conference on IEEE Industrial Electronics Society． Piscataway: IEEE，2011: 2227 ［9］ Pietikainen M，Ojala T，Xu Z． Ｒotation-invariant texture classifi￾cation using feature distributions． Pattern Ｒecognit，2000，33 ( 1) : 43 ［10］ Zhang L，Chu Ｒ F，Xiang S M，et al． Face Detection Based on Multi-block LBP Ｒepresentation． Springer Berlin Heidelberg， 2007: 11 ·1733·