Detection of fabric defects using fuzzy decision tree

This paper presents the representation of fabric texture by four features: Local Binary Patterns (LBP), Local Directional Patterns (LDP), Scale Invariant Feature Transform (SIFT) and Speeded up Robust Features (SURF). The features extracted by these approaches are used in the Fuzzy Decision Tree (FDT) to detect defects in fabrics. We employ both fuzzy Gini index and fuzzy Shannon entropy as the splitting criteria. Two membership functions: Gaussian and trapezoidal are employed for the fuzzification of the genuine and imposter scores. A stopping criterion is devised to terminate the FDT. It is found that LDP features outperform LBP, SIFT and SURF features in the classification of defects in fabrics.