DSP Laboratory - DIEI

This paper describes a simple and effective algorithm for texture defect detection in uniform and structured fabrics. The proposed approach is articulated in two phases: feature extraction and defect identification. The texture features extraction phase relies on a complex symmetric Gabor filter bank and Principal Component Analysis for dimensionality reduction. Opposite to most previous works, our analysis is performed on a patch basis, which has shown to be more effective than simply considering raw pixels as features. The defect identification phase is very fast as it is based on evaluating the Euclidean norm of the patch feature vectors and comparing it with fabric type specific parameters. A calibration procedure, performed offline, is adopted in order to estimate the optimal parameters. The performance of the algorithm has been extensively evaluated on a publicly available image database. The results show that, despite its simplicity, our algorithm outperforms previous approaches in most of the considered cases, achieving a detection rate of 98.8% and a false alarm rate as low as 0.37%.