基于频域互相关的LED晶粒阵列快速视觉检测  

Rapid Vision Measurement of LED Die Array Based on Cross Correlation in Frequency Domain

作　　者：李鹤喜[1] 韩新乐 杨铁牛[2] LI He-xi;HAN Xin-le;YANG Tie-niu(School of Computer Science, Wuyi University, Jiangmen, Guangdong 529020, China;School of Mechanical and Electrical Engineering, Wuyi University, Jiangmen, Guangdong 529020,China)

机构地区：[1]五邑大学计算机学院,广东江门529020 [2]五邑大学机电工程学院,广东江门529020

出　　处：《计量学报》2018年第4期476-480,共5页Acta Metrologica Sinica

基　　金：国家科技重大专项(2011ZX02403-004);广东省自然科学基金(2016A030313003);江门市科技计划项目(20140060117111)

摘　　要：为了满足LED晶粒视觉检测的实时性要求,提出了一种频域互相关的快速模板匹配方法。首先通过一种递推求和算法将LED晶粒阵列图像的灰度在空间域进行归一化处理,然后利用快速傅里叶变换到频率域进行模板与LED晶粒阵列图像的互相关运算,快速定位所有LED晶粒的像素坐标。对于清晰的二值晶粒图像,可以省去归一化处理,检测效率更高。实验结果表明:对于复杂结构的LED晶粒,检测效率远高于传统的NCC模板匹配,比常用的序贯相似检测(SSDA)算法效率提高1倍以上。该方法能够用于各种LED分选或固晶系统中的晶粒检测。A rapid template matching method based on cross correlation in frequency domain（CCFD） is presented for real-time visual measurement of light-emitting diode （LED） die. First, the gray value of LED die array image is normalized by a recursion-sum algorithm in spacial domain, and then using fast Fourier transform the CCFD is carried out between LED die array image and die template to rapidly locate all pixel coordinates of LED dies. The higher measurement efficiency can be obtained because of omitting image normalization for clear binary images of LED die array. Experimental results of complex LED die images show that the measurement efficiency is far higher than conventional normalized cross-correlation （NCC） algorithm, and is only half of consumed time to compare with the sequential similarity detection algorithm（ SSDA）. The proposed CCFD algorithm can be used for rapid LED die measurement in LED sorting or die-bonding system.

关 键 词：计量学 LED晶粒 视觉检测 频域互相关 模板匹配 

分 类 号：TB96[机械工程—光学工程]