Skip to Main Content
An approach to identify 7 types of common defects in silk fabric by combining wavelet transform, generalized Gaussian density (GGD), defect segmentation and learning vector quantization (LVQ) neural network is proposed in this paper. 350 fabric defect images of 7 different types, including non-defect ones, 50 images of each type, are decomposed at three different levels with wavelet base , coif4, and wavelet coefficients in each subband are independently modeled by GGD, while the scale and shape parameters of which are extracted as textural features. To describe the characteristics of defect fully, the geometrical feature, the ratio of max length Lmax and max width Wmax, is also extracted from the segmented defect image using the optimal threshold segmentation algorithm. For comparison, two energybased features are also extracted as textural features from wavelet coefficients directly, the number of which is the same as the scale and shape parameters estimated from GGD model with maximum likelihood (ML) estimator. Experimental results on the 350 fabric defect images indicate the proposed method is realizable and successful, especially when each fabric defect image is decomposed at level three, 18 textural features extracted from the GGD model and 1 geometrical one calculated from the segmented image, these 19 features of every sample are used to train and test LVQ neural network, the average identification accuracy of 7 types defects is 99.2%.