关键词: 机织物, 瑕疵检测, 奇异值分解, 基向量, 重构误差, AR模型

Abstract: Aiming at texture adaptability and real time issue challenging for most existing algorithms, a method for woven fabric defect detection using singular value decomposition (SVD) is proposed. Firstly, the gray values of the normal image are projected along the horizontal and vertical directions, and the resulting two sequences are combined into a joint sequence. Secondly, the matrix of the joint sequence is solved by SVD, extracting basis vectors. Finally the extracted basis vectors are used to reconstruct testing samples and its reconstruction error can be used to discriminate defects from normal texture. The effect of the number of vectors and patch size is also investigated. After 2888 experimental samples ,results show that False detection rate of less than 10% and the detection rate of more than 90%. And the proposed method outperforms AR method in detection accuracy and real time.

Key words: woven fabric, defect detection, singular value decomposition, basis vector, reconstruction error, AR model