Visual uniformity recognition of nonwovens with contourlet energy features and K-NN classifier

Abstract

The surface uniformity recognition and inspection is an important and vital procedure of nonwoven production. Here, a novel approach for the uniformity recognition of nonwovens based on contourlet energy features and K-NN classifier is proposed. In this paper, the uniformity recognition based on contourlet transform and K-NN classifier is considered as a special case of pattern recognition problem that will be solved in two stages. In the first stage, the nonwoven images are decomposed with contourlet transform and then two energy features, norm-1 L1 and norm-2 L2 are calculated from contourlet coefficients of each bandpass directional subband. In the second stage, the extracted energy-based features are used to train and test K-NN classifier, and for comparison, the experimental results coming from different feature set, L1, L2 and L12 (the combinations of L1 and L2) are discussed. In experiment, when the nonwoven images are decomposed at level 3 using contourlet transform and 24 energy-based features, i.e., L12s, are used to train 1-NN classifier, the average recognition accuracy is 98.4 %, which is superior to the method based on wavelet energy-based features.