摘要
针对化妆品纸质标签生产中出现的不干胶标贴漏贴、偏移、倾斜、叠加,生产日期字符漏喷、偏移、倾斜、多喷,日期喷印错误等缺陷,提出一种化妆品纸质标签缺陷视觉检测方法。首先,利用数字形态学与连通域分析技术,从倾斜校正后的标签本体图像中分别提取不干胶标贴和日期字符区域;其次,分别提取区域重心坐标与方向角检测不干胶标贴和日期字符的位置缺陷;进而,在日期点阵字符垂直校正基础上分割字符,提取网格特征、垂直投影、投影宽度及字符占空比等字符特征,利用这些特征训练BP神经网络并实现对日期点阵字符的识别;最后,在样机上采集了大量正常及具有不同类型缺陷的化妆品标签图像,进行实验验证。实验结果表明:该算法准确率高、稳定性好,能够快速检测位置缺陷和准确识别日期字符,检测准确率可达94.4%。
In view of the defects in the production of cosmetic paper labels, such as the leakage, offset, tilt and overlay of adhesive sticker labels as well as the leakage, offset, tilt, multiple spraying and date printing errors of the production date characters, this paper proposes a visual inspection method for the defects of cosmetic paper labels. Firstly, by using the digital morphology and the connected component analysis technology, we extracted the adhesive sticker label and date character area respectively from the image of the body after tilt correction. Secondly, in order to detect the position defects of the label and date characters, the coordinates and directional angles of the center of gravity of the region were respectively extracted. Thirdly, on the basis of vertical correction of the dot matrix characters of date, characters were segmented to extract grid features, vertical projection, projection width and character duty ratio, and these features were used to train the BP neural network and identify the date dot matrix characters. Finally, a large number of normal cosmetic label images and those with different types of defects were collected on the prototype for experimental verification. Experimental results showed that the algorithm had high accuracy and good stability. It could not only quickly detect the position defects, but also accurately identify the date characters, and the detection accuracy could reach 94.4%.
作者
刘鹏
戴文战
LIU Peng;DAI Wenzhan(Institute of Automation,Zhejiang Sci-Tech University,Hangzhou 310018,China;School of Information and Electronic Engineering,Zhejiang Gongshang University,Hangzhou 310018,China)
出处
《浙江理工大学学报（自然科学版）》
2019年第2期231-238,共8页
Journal of Zhejiang Sci-Tech University(Natural Sciences)
基金
国家自然科学基金项目(61374022)
浙江省基础公益研究计划项目(LGG18F030001)
金华市科学技术研究计划重点项目(2018-1-027)
关键词
纸质标签
质量检测
特征提取
点阵字符识别
机器视觉
paper label
quality detection
feature extraction
dot matrix character identification
machine vision
