基于光照预处理与特征提取的纺织品瑕疵检测方法

针对光照对纺织品图像特征提取的影响以及传统完整局部二值模式(complete local binary pattern)算法的局限性,本文提出了一种基于局部对比度增强(local contrast enhancement)算法的改进CLBP特征提取方法并将其应用到纺织品瑕疵检测中。该方法采用局部对比度增强算法对受光照影响的纺织品图像进行预处理,使用改进CLBP算法对分块后(格分割)图像进行特征提取,计算每一格子特征值与标准特征值的KLD散度并与训练得到的阈值进行比较,大于阈值格子标记为瑕疵。使用本文方法在标准星形(star)数据库与箱形(box)数据库中实验结果表明,该方法与其他预处理方法相比有更加出色的处理效果,大部分检测结果的查全率均可达到0.99左右。

Fabric defect inspection based on illumination preprocessing and feature extraction

In this study, a modified complete local binary pattern (CLBP) feature extraction method based on the local contrast enhancement is proposed to address the influence of illumination on the feature extraction of fabrics and the limitations of the traditional CLBP algorithms. In this method, a local contrast enhancement algorithm is used to preprocess the fabric images that are affected by illumination, and the modified CLBP algorithm is used to calculate the feature value of every lattice in every image after lattice segmentation and to subsequently calculate the relative divergence KLD between the feature of every lattice and the standard feature value. Subsequently, the calculated relative divergence will be compared with the threshold, and the lattice that exhibits a value larger than the threshold is considered to be a defect area. The experimental results obtained using the standard star pattern and the box pattern databases denote that the proposed method is better than the remaining preprocessing methods, and the recall rate of the majority of the detection results obtained using this method can reach approximately 0.99.