基于图像处理技术的树脂镜片瑕疵分类特征研究

在基于机器视觉技术的镜片自动检测系统中,为了实现镜片的分级,需要对镜片瑕疵进行分类。利用建立的图像获取系统自动检测树脂镜片瑕疵;针对镜片点杂质、划痕和羽毛3种瑕疵的分类问题,提出了以圆形度、直线拟合相关系数作为分类特征的方法,通过数据统计计算出分类特征的阈值,并分析了细化算法对划痕和羽毛直线拟合相关系数的影响,结果表明,细化算法对于提高划痕和羽毛的分类准确率具有重要作用。实验验证了算法的可行性,并分析了误判的原因。实验结果表明,3种瑕疵的分类准确率为96%。

Resin lens defect classification based on image processing

During the lens automatic detection ba sed on machine vision technology,the defects should be classified in order to ac hieve the grading of the lenses.Automatic detection system components, working principle and the resin lens major defects are simply introduced.As for the classification of the three main defects of point impurities,scratches and feathers,the circularity and straight-line fitting correlation have been propos ed in this paper as the classifying features.The threshold of the classifying features has been calculated on the basis of data analysis and statistics.The i nfluence of the thinning algorithm on the straight-line fitting correlation has been discussed for the defects of scratches and feathers.The experimental results show that the thinning algorithm plays an imp ortant role in improving classification accuracy of scratches and feathers.The experiment is conducted to analyze the feasibility of the algorithm.The results indicate that this method has a high c lassification accuracy of 96% for the three defects,which meets the requiremen ts of the detection system.