赵景波+赵建辉+许晓帆
摘 要: 为了降低引线框架缺陷识别的误检率,比较了基于灰度的模板匹配和基于特征的匹配算法之间的优缺点,并针对引线框架缺陷检测中参考图和检测图存在差异的特点,提出基于区域定位和不变矩的特征匹配算法。该算法通过边缘检测定位出特征区域,并用不变矩进行区域特征描述。在缺陷识别试验中,相比于模板匹配算法,该算法表现出更快的结算速度,更高的配准精度,更低的配准失败概率。结果表明,该配准算法适用于引线框架缺陷识别,降低误检率。
关键词: 引线框架; 机器视觉; 模板匹配; 特征匹配算法
中图分类号: TN911.73?34 文献标识码: A 文章编号: 1004?373X(2014)04?0101?03
Image registration algorithm for lead frame defect detection
ZHAO Jing?bo1, ZHAO Jian?hui2, XU Xiao?fan1,2
(1. Northwestern Polytechnical University, Xian 710129, China; 2. Unit 93995 of PLA, Xian 710306, China)
Abstract: To reduce the false detecting rate in lead frame defect recognition, the advantages and drawbacks of template matching based on gray and matching algorithm based on invariant moment. A feature matching algorithm based on area position and moment invariant is proposed according to the characteristic that the difference exists in conference image and sensed image during lead frame defect detection. The algorithn can ensure the feature area location by edge detection and describe the region feature by means of invariant moment. In the test of defect recognition, the algorithm showed up more rapid calculating speed, higher registration accuracy and lower registration false rate than the template matching algorithm. The results show that the image registration algrithm is suitable for lead frame defect detection and reduction of the false detecting rate.
Keywords: lead frame; machine vision; template matching; feature matching algorithm
引线框架作为集成电路的芯片载体,是一种借助于键合材料(金丝、铝丝、铜丝)实现芯片内部电路引出端与外引线的电气连接,形成电气回路的关键结构件[1]。对于集成度高的芯片,其引线框架采用蚀刻工艺制造。在工艺实施过程中,产品中会出现一些特有的缺陷[2]。目前,部分生产企业采用了基于机器视觉的缺陷检测系统,但是,此类检测系统普遍存在误检率过高的问题,导致产量很难提高。本文分析了误检率高的原因,针对图像配准环节提出了改进的算法,可有效改善检测效果。
1 缺陷检测流程
1.1 图像预处理
对于由光学传感器获得的数字图像,由于成像条件、光照条件及采集设备等的影响,往往使所采集图像中的有用信号淹没在噪声中。因此,图像预处理通常包括噪声去除滤波、二值化、图像边缘检测等几个方面,使图像特征更加明显[3]。……