Decision Fusion for Visual Inspection of Textiles

Defect detection of textiles is a necessary requirement for quality control and customer satisfaction. This paper presents a system for decision fusion in order to enhance the accuracy of defect detection in textiles. A multi-classifier decision fusion technique based on majority voting is presented to solve the problems of sensitivity to parameter variation and to make use of the advantages of the individual feature sets for accurate texture characterization. Features based on Gray Level Co-occurrence Matrix (GLCM), Laws Energy (LE) Filter, Local Binary Patterns (LBP), HU Moment invariants, Moment of Inertia (MOI) and Standard Deviation of Gray levels are used to train a set of Learning Vector Quantization (LVQ) classifiers. Detection accuracies of classifiers trained on single-feature sets are compared with those of the majority voting among the individual classifiers. The results obtained from majority voting indicate that the decision fusion technique improves the accuracy and reliability of the detection process. Empirical results indicate the high accuracy of the presented approach. The correct defect detection rate of the proposed approach is 98.64% with an average false acceptance rate of 0.0012.