The exisiting fabric defect detection methods based on image processing is poor in real-time performance and low in accuracy. In order to solve this problem, an algorithm consisting of two stages of learning and detection was proposed. By means of learning the gradient energy features and their distribution properties of non-defect model images, parameters in the detection stage were obtained automatically. On the one hand, by using integral image theory, summation operation in the image patch with arbitrary size was simplified to three addition operations, and gradient energy features in fabric images were extracted very quickly, so that fabric defects can be detected in real time. On the other hand, kernel functions were used to fit the distribution of feature parameters, mean shift method was used to solve the peak value in the distribution, and an adaptive threshold was obtained, so that fabric defect can be segmented precisely. In the experiments, the proposed algorithm was compared with the other three methods, respectively, based on local binary pattern features, wavelet features and regular band features. Tests on fabric image datasets including three kinds of textures and six kinds of defects show that the proposed method has an average running time 56 ms and the accuracy rate is 97%.
Template