Abstract: Aiming at the difficulty of detecting defects on the surface of complex textured chips,a chip surface defect detection algorithm based on improved Frangi filter is proposed.The improved algorithm breaks through the limitation that the traditional Frangi filter can only effectively enhance linear defects.Based on the original algorithm model,the circular structure suppression term was removed,then the optimized frequency tuned saliency detection algorithm was used to improve the background determination condition,and finally the constructed block defect response term was added to obtain a new multi-feature response function.The experimental results show that the improved algorithm can effectively enhance various types of defects on the chip surface,and the average detection rate reaches 97.3%,which is 46% higher than the original algorithm.Compared with similar detection algorithms,the improved algorithm has obvious advantages in weak scratch detection,and the scratch detection rate reaches 90%.In addition,the average time consumed by the algorithm is only 0.08s,which meets the real-time detection demand.

Key words: chip defect detection, Frangi, multi-feature response, Hessian matrix, saliency detection

摘要： 针对复杂纹理芯片表面缺陷检测困难的问题,提出一种基于改进Frangi滤波的芯片表面缺陷检测算法。改进算法突破了传统Frangi滤波仅能有效增强线状缺陷的限制,在原算法模型的基础上,先将圆形结构抑制项去除,再采用优化后的频率调谐(frequency tuned,FT)显著性检测算法对背景判定条件进行改进,最后加入所构建的块状缺陷响应项,得到新型多特征响应函数。实验结果表明,改进算法能够有效增强芯片表面各类缺陷,平均检测率达到97.3%,较原算法提升46%。与同类检测算法相比,改进算法在弱划痕检测方面优势明显,划痕检测率达到90%。此外,算法平均耗时仅0.08s,满足实时检测需求。

关键词: 芯片缺陷检测, Frangi, 多特征响应, Hessian矩阵, 显著性检测