On-loom fabric inspection using multi-scale differentiation filtering

An approach based on multi-scale differentiation filtering (MSDF) is proposed for fabric defect detection. Multi-scale representations for the first derivatives of fabric images are obtained by MSDF; normal ranges of such outputs are estimated and singularities caused by defects or noises are recognized. Noise is further removed from outputs based upon the difference between the correlations of defect outputs and that of noise outputs. By applying to various cases of defect detections, the performance of the approach is evaluated. It has both the high anti-interference of texture outputs and high sensitivity for defect detection. With fast B-spline algorithms, the computation complexity of the approach is reduced.