Abstract
In this paper, a method is proposed for the fabric defect detection based on the two-level K-Nearest Neighbor classifiers. First, six features are extracted from the directional grey level co-occurrence matrix of the fabric input image. Next, the minimum, maximum, median, and mean of intensities of the input image are calculated. Then, the Principal Component Analysis (PCA) algorithm is applied to reduce the feature vector dimensions. Finally, the first K-Nearest Neighbor (KNN) classifier is used for these features clustering. As a result, the fabric input image is classified to the defective and non-defective based on the trained data. In the second level, the defective fabric image features are extracted and reduced by the PCA and classified by the second KNN. As a result, each defect class is classified and their locations are determined by using the morphological operations. The proposed method performance is evaluated on the TILDA database. The simulation results show more than 90% improvement on the accuracy of the fabric defect detection in comparison to the recent related works.