摘要
提出一种结合小波变换与共现矩阵用于纺织品图像缺陷检测的方法。首先将灰度图像分解成子带;然后将纹理图像分割成互不重叠的子窗口,提取共现特征;最后用无缺陷样品训练的Mahalanobis分类器将每一子窗口划分为缺陷的和无缺陷的。应用该算法进行实际工厂环境中的纺织品缺陷检测。实验结果表明,集中处理具有强判决能力的某一频带提高了检测性能,也改善了计算效率。
The paper put forward a defect-detecting method on texture image accomplished by the combination of wavelet transformation and co-occurrence matrix. First divided the gray image into sub-bands, then segmented texture image into subwindows without mutual superposition, abstracted whose co-occurring features, finally, labeled each sub-window with ＂faulty＂ or ＂flawless＂ by Mahalanobis classifier having undergone flawless sample training. The method of calculation has been applied to the daily practice of textile＇ s defect-detecting in factories, the outcome indicating that the calculation efficiency, as well as detecting capability can be intensified by a certain band of frequency with high discrimination scale.
出处
《计算机应用研究》
CSCD
北大核心
2007年第11期57-59,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(50475044)
广东省自然科学基金资助项目(7007362)
广州市黄埔区科技计划资助项目(0713)
关键词
纹理缺陷分析
共现矩阵
小波滤波器
texture defect analysis
co-occurrence matrix
wavelet filter
