基于机器视觉古陶瓷无损分类识别  

Machine Vision Based Classification and Identification for Non-destructive Authentication of Ancient Ceramic

作　　者：翁政魁 管业鹏[1,2] 罗宏杰[3] 

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]新型显示技术及应用集成教育部重点实验室,上海200072 [3]上海大学材料科学与工程学院,上海200444

出　　处：《硅酸盐学报》2017年第12期1833-1842,共10页Journal of The Chinese Ceramic Society

基　　金：国家自然科学基金(11176016;60872117);高等学校博士学科点专项科研基金(20123108110014)资助

摘　　要：为客观、有效地对古陶瓷进行无损分类,提出了一种基于机器视觉古陶瓷无损分类识别方法。通过遍历古陶瓷器型边缘轮廓,获取古陶瓷器型结构细节特征,并在HSI空间下提取古陶瓷釉色多通道颜色直方图特征。同时,提取反映古陶瓷纹理多样性的LBP纹饰特征。基于上述特征,采用机器学习方法实现古陶瓷器型结构、釉色及其纹饰图案的无损分类识别。结果表明:通过机器视觉可以有效地对古陶瓷进行分类识别;在以16为曲率步长、9为LBP算子分块数时,分别提取古陶瓷结构,纹饰特征有较好的识别精度,其中,基于结构与釉色融合特征相比单一特征具有更好的识别效果;当古陶瓷发生结构或纹饰上的小部分缺损时,该方法可以保持一定的鲁棒性,当信息丢失或缺损为5%时,平均识别率依旧可达85%以上,可期望实现古陶瓷科技鉴定中的良好应用。A classification method was proposed for non-destructive authentication of ancient ceramic based on machine vision to make the ancient ceramic identification more objective and accurate. The edge contour of ancient ceramic was traversed at first. The detail feature for the shape structure of ancient ceramic was determined by the corresponding curvature. A multi-channel color histogram in HSI space was proposed to describe the glaze colors. Meanwhile, the LBP texture feature was extracted to reflect the colorful graphic pattern of ancient ceramic. Machine learning was used to perform classification and identification for non-destructive authentication of ancient ceramic in the features mentioned above. The experimental result shows that the ancient ceramic can be effectively identified by machine vision and set 16 as the curvature step, 9 as LBP operator block numbers during the extraction of structural and glaze feature outperforms than other parameters. Also, the fusion of structural and glaze features obtain a better recognition accuracy, compared to the single feature. The method can maintain a relative robustness when a small part of the ceramic structure or glaze is damaged. The average recognition accuracy can still reach more than 85% when the information loss or defect is 5%, which could allow its application in ancient ceramic scientific identification.

关 键 词：古陶瓷 科技鉴定 机器视觉 结构信息 釉色信息 纹饰特征 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]