时-空域多特征证据学习与增强的印染疵点在线检测

 提出了一种基于时空域多特征证据学习与增强的织物印染疵点在线检测新方法。利用多种类纹理特征在特征表达上的互补性以及可疑图像分块前n帧历史的对应特征,达到多证据印证的特征学习与分类增强,是一种比较通用的表面缺陷实时检测解决方法。检测总体思想是从“已知的”无疵点纹理表面提取特征,根据特征对被测织物进行分类比较,从而检测出“未知的”疵点纹理区域。检测过程分为一次性时空域多特征证据自学习和在线分类检测两阶段。对实际织物图像序列的在线检测显示,对单色织物常见印染缺陷的有效检测速度达到了55帧/s(1 024×393像素分辨率仿真视频图像),动态检出正确率达到95%以上。

On-line detection of the dyed and printed fabric defects by multi-features evidence learning and enhancement in spatiotemporal domain

A novel method of defects detection for dyed and printed fabrics is presented,which is based on multi-features evidence learning and enhancement in spatiotemporal domain.It′s a general solution to many real-time surface inspection issues.The mutual compensation of multi-features is used to enhance the defects evidence,and history information of the doubted patches in video sequence is also applied to help checking out what are the true defects.The main idea is to find out the unknown defects by comparing the extracted surface features of the known defect-free fabric with those of the fabric being examined.This inspection procedure is divided into two stages: one for the roll-style multi-features learning of the known defect-free textile,the other for realtime surface inspection.Many experiment results of the on-line inspection show that the efficient detection speed reaches 55 frames per second to the image sequence((1 024)×393 pixels) for dyed and printed fabrics of single-color,with a correct dynamic check out rate on surface defects above 95%.