Fabric defect classification using wavelet frames and minimum classification error training

This paper proposes a new method for fabric defect classification by incorporating the design of a wavelet frames based feature extractor with the design of a Euclidean distance based classifier. Channel variances at the outputs of the wavelet frame decomposition are used to characterize each nonoverlapping window of the fabric image. A feature extractor using linear transformation matrix is further employed to extract the classification-oriented features. With a Euclidean distance based classifier, each nonoverlapping window of the fabric image is then assigned to its corresponding category. Minimization of the classification error is achieved by incorporating the design of the feature extractor with the design of the classifier based on minimum classification error (MCE) training method. The proposed method has been evaluated on the classification of 329 defect samples containing nine classes of fabric defects, and 328 nondefect samples, where 93.1% classification accuracy has been achieved.