结合全局和随机局部频率调谐的复杂纹理表面缺陷检测  

The defect detection for random texture surface combined with the whole and randomly partial frequency tuned

作　　者：李健[1] 陈海丰[1] 吴华[1] 

机构地区：[1]陕西科技大学电气与信息工程学院,陕西西安710021

出　　处：《陕西科技大学学报（自然科学版）》2014年第2期152-155,163,共5页Journal of Shaanxi University of Science & Technology

基　　金：陕西省教育厅专项科研计划项目(11JK1065);温州市科技合作项目(H20090045)

摘　　要：针对复杂纹理表面缺陷检测过程中缺陷区域大小不一、受自身纹理干扰大等问题,提出了一种将全局频率调谐和随机局部频率调谐相结合的缺陷检测新方法.该方法首先对整幅图像进行频率调谐以得到全局显著图;然后对随机局部分块图像进行频率调谐以得到局部显著图;接着将这两类显著图相融合得到综合显著图;在此基础上,通过对综合显著图进行阈值分割以实现对缺陷区域的检测.实验表明,该方法可以抑制自身纹理,同时完整突出大小不一的缺陷目标,在提高检测准确度的同时,没有降低检测的实时性,与传统方法相比有较强的优越性.This paper proposes a new defect detection method ,which combines the whole fre-quency tuned with the randomly partial frequency tuned ,to solve the problem of the different defect size ,the strong interference by its own texture arising in the course of defect detec-tion .By taking the frequency tuned of the whole image and the random partial blocks ,we ob-tained the combination of the global significant map and the partial significant map ,named integrated saliency map .Segmenting threshold of this map can achieve the detection for de-fect areas .Experiment show s that this method has the better real-time .Not only can it re-strain the texture ,but it is able to protrude defect targets as well .Compared with the tradi-tional methods ,it has certain superiority .

关 键 词：机器视觉 缺陷检测 频率调谐 显著图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]