摘要
针对布匹图像非下采样Contourlet分解系数能更好地描述瑕疵图像的轮廓特性,同时具有平移不变性和多方向性等优点,提出一种新的瑕疵自动检测算法.该算法通过非下采样Contourlet变换得到图像的多尺度、多方向稀疏表示;在此基础上,通过代价函数选择最优子带,得到较鲁棒性的描述;最后实时地估计瑕疵和非瑕疵图像的混合高斯模型参数,有效地避免了对每一类瑕疵的估计,显著地降低了计算量.实验结果表明,与现有经典算法相比,该算法的主观效果和客观评价性能都有明显改进.
Considering the advantages that decomposition coefficients in the non subsampled Contourlet of fabric images can describe the contour characteristics in a better way, and that they have shift-invariant and multidirection, a novel algorithm for automated detection of fabric defect images is presented. Firstly, the nonsubsampled Contourlet transform （NSCT） is used to perform the sparse representations in multi-scales and muhi-directions. On this basis, the optimal sub-bands of NSCT are selected by the cost function, and then the robust descriptions are obtained. Finally, the parameters of defect and nondefect images are timely estimated separately by the Mixture Gaussian Model（MGM）, which effectively avoids estimating each defect and reduces the computational complexity evidently. Experimental results show that the proposed algorithm can lead to a better performance than the traditional algorithms in subjective effects and objective evaluation.
出处
《西安电子科技大学学报》
EI
CAS
CSCD
北大核心
2011年第5期65-72,共8页
Journal of Xidian University
基金
国家自然科学基金资助项目(60872141)
陕西省自然科学基础研究计划资助项目(2009JQ8019)
综合业务网理论及关键技术国家重点实验室基金资助项目(ISN090302)
西安电子科技大学基础科研业务费资助项目(K50510010007)
