低信噪比小检测目标光学薄膜缺陷图像分割算法研究

bomoaa

低信噪比小检测目标光学薄膜缺陷图像分割算法研究   　
9 b. L) y. D: h' n" `0 I7 A


% |% K+ q) V+ u" V, Q7 Q 【英文篇名】         Low-SNR and small target image segmentation algorithm for defect inspection of optical films
5 \7 U  _% _, K9 I 【作者】        胡广华; 钟球盛;
【英文作者】        HU Guang-hua1; 2; ZHONG Qiu-sheng1（1.School of Mechanical & Automotive Engineering; South China University of Technology; Guangzhou 510641; China; 2.Guangzhou Nansha Huazhuo Chemicals Co.; Ltd.; Guangzhou 511458; China）;
【机构】        华南理工大学机械与汽车工程学院; 广州南沙华卓化工有限公司;
4 X# V: D1 {4 b; y( I7 B) O" r& R 【关键词】        图像分割; 误差修正; 光学薄膜; 缺陷检测;
【英文关键词】        image segmentation; error correction; optical films; defect inspection;
8 C# S( g# }2 ^ 【中文刊名】        计算机应用研究, 2012年 04期
2 Z7 {( t; _6 d: }1 V2 F+ A 【英文刊名】         Application Research of Computers
【中英文摘要】         结合测量误差分析理论,提出一种新的高效的图像分割算法,将薄膜的连续图像采样过程看做等精度测量过程,背景图像认为仅存在随机误差,缺陷像素视做粗大误差。算法在起始阶段通过连续多次采样建立标准背景图像,并统计各像素位置的灰度标准差,然后将实时待测图像中的每个像素与标准背景像素进行减法运算,以3σ作为检测判据,从而将缺陷的检测与分离等效为粗大误差的判别与剔除问题;同时,考虑缺陷像素点的空间分布特征,将上述过程扩展到像素邻域范围,以二维分割阈值取代一维阈值,对离散噪声有明显的抑制作用。实验结果表明,该算法运算速度快,对低对比度图像适应性好,对噪声不敏感,可实现对不同种类缺陷的准确、完整分割,适于实时光学薄膜检测应用。
Based on error correction theory,this paper proposed a novel algorithm to inspect defects rapidly.With an assumption that only random errors existed in a background image in which the random errors followed a normal distribution,the defect inspection problem converted to a gross error correcting problem,which could be solved naturally by using the 3σ rule.The algorithm applied 2D thresholds generated from neighbor pixels to reduce the influence of noise.A comparison of the computational results with the exi...