摘要
为了降低智能手机PCBA电路板元件反光特性对图像灰度产生的干扰,提高边缘连接的完整度,减少伪边缘的数量,本文对传统的canny算子进行改进,用于提取智能手机PCBA电路板的边缘信息。首先,兼顾滤波器去噪与梯度保持特性,使用改进的具有“动态”惩罚因子的引导滤波器替代高斯滤波器,减少边缘点的丢失。然后,使用近邻域局部自适应阈值法取代传统双阈值法,用于解决元件稠密区域存在灰度变化频繁背景与目标差异较小而导致阈值分割不准确的问题。实验结果表明,当滑动窗口保持在19大小时,具有最好的阈值分割效果,伪边缘数量降到最低。本文方法处理后的图像,拥有更完整细化的边缘,伪边缘数量大大减少,局部元件稠密区域,边缘细节得到保留,符合像素级手机PCBA电路板的边缘检测精度要求。
To reduce interference to the grayscale of an image due to the reflective characteristics of the PCBA components in a smartphone,improve the integrity of the edge connection,and reduce the number of false edges,this paper proposes an improvement in the traditional Canny operator to extract the PCBA from smartphone edge information.First,considering the characteristics of filter denoising and gradient retention,the use of improved guide filters with“dynamic”penalty factors instead of Gaussian filters reduces the loss of edge points.Thereafter,the traditional double threshold method is replaced with the nearest neighbor local adaptive threshold,which is used to solve the problem of inaccurate threshold segmentation due to frequent grayscale changes in the dense area of the component and the small difference between the background and the target.Experimental results show that a sliding window of 19 has the best threshold segmentation effect,and the number of false edges is minimized.The image processed by this method has more complete and refined edges,the number of false edges is greatly reduced,and the dense local-component area and edge details are retained,which meets the edge detection accuracy requirements of pixel-level mobile phone PCBAs.
作者
李彦
赵其峰
闫河
李晓玲
谢敏
LI Yan;ZHAO Qi-feng;YAN He;LI Xiao-ling;XIE Min(Computer Science and Engineering, Chongqing University of Technology, Chongqing 401320, China;Liangjiang Artificial Intelligence Academy, Chongqing University of Technology,Chongqing 401147, China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2020年第9期2096-2102,共7页
Optics and Precision Engineering
基金
国家重点研发计划“智能机器人”重点专项项目资助(No.2018YFB1308602)
国家自然科学基金面上项目资助(No.61173184)
重庆市自然科学基金资助项目(No.cstc2018jcyjAX0694,No.cstc2018jcyjA2328)。
