基于低秩分解的织物疵点检测

Detection for fabric defects based on low-rank decomposition

针对传统低秩分解法导致的图像信息过度丢失和织物弹性导致的歪斜问题,提出一种基于Beta范数的改进低秩分解检测方法。首先,通过提取织物图的基元特征构造先验信息图。其次,采用Beta范数代替低秩分解中的核范数,并由先验信息图引导低秩分解方法对织物图进行分解,解决了传统低秩分解方法中核范数导致的图像信息过度丢失的问题。进而,提取织物图的方向梯度直方图(HOG)特征构造后验信息图,并将后验信息图和通过低秩分解得到的稀疏分量进行哈达玛乘积获得显著图,解决了织物弹性导致的歪斜问题。最后,利用最优阈值分割得到疵点图。将实验结果与已有的4种方法进行对比,结果表明,该方法可以有效抑制歪斜干扰,且检测时间更短。

Aiming at excessive loss of image information in fabric defect detection caused by the commonly used low-rank decomposition method and the weft skew caused by fabric elasticity, an improved low-rank decomposition detection method based on Beta norm was proposed. This method starts by constructing a prior map by extracting the texton feature of the fabric image. Second,a Beta norm was used to replace the nuclear norm in the low-rank decomposition process, whereas the low-rank decomposition was guided by the prior map to decompose the fabric image. Compared with the nuclear norm,it was found that the proposed method does not lead to excessive loss of image information.Furthermore,a posterior map was constructed by extracting the HOG( histogram of oriented gradients)feature of the fabric image,and a saliency map was obtained by the Hadamard product between the posterior map and the sparse component obtained by the low-rank decomposition which can solve the skew problem caused by fabric elasticity. Finally,optimal threshold segmentation was used to obtain the defect figure. Compared with the existing four methods,the experimental results demonstrate that the proposed method can effectively suppress the skewness in the fabric and the detection time is shorter.