Abstract: Accurate detection of glue painting defects is of great significance for improving the quality of glue painting process of micro camera module.For this reason,a method for detecting the defect of painting glue in micro camera module was proposed.The glue images obtained by specific circle light source and 3CCD camera were analyzed,and a glue image segmentation method with adaptive multiple thresholds was proposed.On this basis,the image features of six ordinary types of painting glue such as good product,overflow,deficiency,breakage,scratch and offset were analyzed within painting processes,the global and local features of painting glue images were extracted,and the detection representation model and the detection method of these six types of painting glue was established.The detection results based on the proposed method within 1059 collected glue images were obtained and evaluated by Misclassification Error,IntersectionOverUnion and Dice value.The results illustrated that the accuracy of the proposed algorithm for glue defect detection could reach 95.94% while the missed kill rate was 0.38%,while the time cost of the detection was within 0.15s,which verified that the proposed algorithm had sufficiently high detection precision and effectiveness.

Key words: micro camera module, painting glue, image segmentation, defect detection

摘要： 为准确检测画胶缺陷以提升微型相机模组画胶工艺质量,提出一种微型相机模组画胶缺陷检测方法。通过分析由环形光源和3CCD相机获取的画胶图像特征,提出一种自适应多阈值画胶图像分割方法,在此基础上分析画胶工艺过程中常见的良品、多胶、少胶、断胶、刮胶、偏移6种类型的画胶图像特征,提取画胶图像的全局特征和局部特征,并建立6种画胶类型特征的矩阵表达模型与检测方法。最后采集1059张画胶图像进行缺陷检测实验,并采用错误分类误差、交并比和骰子系数对检测结果进行评估。实验结果表明,所提方法对画胶缺陷检测的正确率为95.94%,漏杀率为0.38％,同时单次检测时长小于0.15s,检验了所提方法的准确性和时效性。

关键词: 微型相机模组, 画胶, 图像分割, 缺陷检测