An Improved LBP Method for Feature Extraction and Classification of Metal Defects Based on Gabor Filter and 2DPCA

摘要

Metal is widely used in all aspects of industrial production and daily life. However, in the process of metal production, preservation, transportation and so on, it is inevitable to lead to different types and shapes of defects on the metal surface. In order to detect defects, first of all, we need to classify the defects, so we need to extract sufficient metal samples for feature extraction, from which we can extract effective attributes or information about defects. Then the extracted features are selected to reduce the error of defect classification. In the metal surface defect detection system, defect recognition is one of the key steps, which belongs to the multi classification problem. LBP has a good description ability when the dimension of defect image is low, but it is not suitable for high-dimensional complex feature extraction. Therefore, this paper extracts the improved method: firstly, Gabor filter is used to denoise and enhance the edge information of the defect, then LBP is used to extract the histogram vector of the amount defect image, and 2DPCA is used to reduce the dimension of the initial data, then SVM is used as the classifier to classify the metal surface defects. Experimental results show that this method is more accurate than LBP and SVM.