刘锐 张长胜 张汉平

摘要:图像分类识别技术主要处理彩色或灰度图像,使用图像特征提取方法提取出同类样本的共性及不同样本的差异性,再配合恰当的分类器设计得到較高的分类识别率。选取200个锡基合金,通过二值化、形态学边缘提取和扫描线种子填充算法寻找封闭区域,用改进的多边形包含算法进行封闭区域包含判定,提取出金相组织的形状特征。对识别的金相组织设计双阈值级联分类器,根据特征识别进行归类。实验结果表明,所设计的分类器测试精度达到96%,识别速度快,能满足实际需求。
关键词:图像分类;多边形包含;边缘提取;特征识别;双阈值级联
DOIDOI:10.11907/rjdk.172823
中图分类号:TP317.4
文献标识码:A 文章编号文章编号:1672-7800(2018)008-0212-04
英文摘要Abstract:The processing object image classification technology mainly deals with color or grayscale images,using image feature extraction method to extract similar samples properly,similarities,differences in the different samples,the classification with the appropriate classifier design is finally able to get a higher rate; 200 tin alloy were selected through the binarization.Morphological edge extraction and scan line seed fill algorithm were employed to find the closed areas which include judging by the improved algorithm of polygon.The shape features of the metallographic structure are extracted.A double threshold cascade classifier is designed to classify the recognition tissue according to the feature recognition.Experimental results show that the accuracy of the proposed classifier is 96%.The recognition speed is fast and the actual demand is met.
英文关键词Key Words:image classification ;polygon inclusion;edge extraction;trick recognition ;dual threshold cascade
0 引言
数字图像中的形状特征提取,大量运用于定位识别[1]、果类自动分级系统[2]、果类自动采摘[3]中,通常从颜色、形状、纹理3个特征进行提取[4],用轮廓特征和区域特征表征形状特征[5]。
一般用提取对象的边界特征值表征轮廓特征,提取对象的整个区域特征表征区域特征。文献[6]采用分支限界法对对象的骨架图进行匹配提取区域特征。文献[7]利用径向基网络成功实现遥感影像的识别分类。上述方法效率高,但计算量大。文献[8]利用SUSAN算子识别图像中角点信息,该方法适用于特征单一对象,对特征算子的要求比较高。
本文运用图像阈值分割、边缘提取等方法有效地识别出锡基合金金相图中各相的形状特征,并采用改进方法识别金相图中的包含关系,对被包含部分进行区分,最终获得金相组织的形状特征。……