A Machine Learning-Based Multi-feature Extraction Method for Leather Defect Classification

Automatic inspection for detecting defects in leather is an inevitable task for grading the leather. Researchers from different parts of the globe have developed many leather defect classification models to address the problems of manual inspection. Discriminating defective and non-defective patterns in the leather substrate are challenging due to the inherent texture variations. Performance of the feature extraction and classifier plays a vital role in the recognition of the relevant patterns. Histogram of oriented gradients (HOG) and grey-level co-occurrence matrix (GLCM) along with Hu moments and HSV are implemented to extract the features from the leather images. The pivotal process is the extraction of these local and global features from the leather images. To detect and classify various leather defect types efficiently, a multi-feature algorithm that combines GLCM and Hog features is also investigated. Leather defect classification is performed using linear regression (LR), linear discriminant analysis (LDA), K-nearest neighbour (kNN), classification and regression tree (CART), random forest (RF), support vector machine (SVM) and multi-layer perceptron neural network (MLP). Experimental results show that the highest classification accuracy (89.75%) is achieved using GLCM along with Hu moments, HSV colour features and random forest classifier.