CAJ | 학술논문

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。
침대복잡문리표면결함검측과정중결함구역대소불일、수자신문리간우대등문제，제출료일충장전국빈솔조해화수궤국부빈솔조해상결합적결함검측신방법。해방법수선대정폭도상진행빈솔조해이득도전국현저도；연후대수궤국부분괴도상진행빈솔조해이득도국부현저도；접착장저량류현저도상융합득도종합현저도；재차기출상，통과대종합현저도진행역치분할이실현대결함구역적검측。실험표명，해방법가이억제자신문리，동시완정돌출대소불일적결함목표，재제고검측준학도적동시，몰유강저검측적실시성，여전통방법상비유교강적우월성。
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 .