Feature extraction algorithm based on adaptive wavelet packet for surface defect classification

This paper proposes a feature extraction method to effectively handle the textural characteristics in images with defects in cold rolled strips. An adaptive wavelet packet scheme is developed to produce the optimum number of features automatically through subband coding gain. Also four classical entropy features in the images with defects are used as local features in the spatial domain. A neural network is used to classify the defects from these features. Experiments with real image data show good training and generalization performances of the proposed method.