基于RPCA的皮革图像缺陷检测

提出了一种皮革视觉缺陷检测算法.通过分析皮革图像的低秩特征,将皮革图像缺陷检测问题转变为从低秩背景图像中分离稀疏矩阵图像.首先采用Gaussian高通滤波器对图像进行了预处理,然后利用鲁棒性主成成分分析(RPCA)对图像进行低秩稀疏分解,并采用效率较高的非精确增广拉格朗日乘子法(IALM)求解.对分解后的稀疏图像进行了后处理,最终在二值图像中获得缺陷的形状和位置.该算法的效率及准确率已经在实验中进行了验证,并与现有算法进行了比较.实验表明,该算法可以用来检测各种不同种类和大小的缺陷,检测准确率高且能够提供完整的缺陷掩模.

Leather Image Defect Detection Based on RPCA

This paper proposes a method of leather visual defect detection. By analyzing the low-rank features of leather images, the problem of leather image defect detection is transformed into separating sparse matrix images from low-rank background images. First, Gaussian high-pass filter is used to pre-process the image, then the Robust principal component analysis (RPCA) is used to perform low-rank sparse decomposition on the image, and the Inexact Augmented Lagrangian algorithm (IALM) is used to solve the problem. The decomposed sparse image is post-processed, and finally the shape and position of the defect are obtained in the binary image. The efficiency and accuracy of this method have been verified in experiments and compared with existing methods. Experiments show that the algorithm can be used to detect various types and sizes of defects, with high detection accuracy and a complete defect mask.