LBPV算法在织物瑕疵检测中的应用

Research on Application of LBPV Algorithm in Defect Detection of Fabrics

织物瑕疵检测中提取特征是至关重要的,局部二值模式能提取纹理的局部信息,且具有旋转不变性,对光照不敏感的优势。但是在检测经向、纬向分布的线状瑕疵时,瑕疵样本与正常样本的区分不够明显。为了提高LBP算法的性能,提出基于LBPV模式的瑕疵检测算法。融合图像局部区域的对比度信息,将局部区域LBP微模式的权值设置为局部区域的方差,提取图像的LBPV特征向量。同时,根据织物纹理具有周期性和方向性的特点,设计了频域滤波器,消弱正常纹理的频谱信息,突出了疵点信息,方便算法实现疵点的检测。实验表明,基于LBPV模式的检测方法检测正确率达到90%以上,具有实用价值。

It is vital to extract features in defect detection of fabrics. Local binary pattern can extract local information of texture and has advantages of rotational invariance and insensitivity to illumination. However, defective samples and normal samples are not obviously differentiated in the detection of linear defects distributed in warp and weft directions. To improve the performance of LBP algorithm, this paper puts forward a defect detection algorithm based on LBPV mode; integrates the contrast information in local areas of images, sets the weight of LBP micro mode in local areas as the variance of local areas and extracts LBPV feature vector of images; meanwhile designs frequency domain filter according to periodicity and directivity of fabric texture to weaken the frequency spectrum information of normal texture and highlight defect information for the convenience of realizing defect detection. The experiment shows that the detection method based on LBPV mode has an accuracy over 90% and practical value.