Defect detection in textured materials using optimized FIR filters
In this work, a new approach for defect detection using linear FIR filters with optimized energy separation was developed. The major drawback with many previous approaches is that they are not sensitive enough to detect defect that produce any subtle intensity transitions and consequently cannot guarantee 100% inspection. Furthermore, these approaches require statistical computations (e.g. mean and standard deviation) for their on-line implementation, which makes them computationally complex and requires additional hardware. The approach presented here does not require any on-line statistical computations and is geared for defects that produce very subtle intensity transitions in acquired images.
The Finite Impulse Response (FIR) filter that guarantees optimal discrimination of energy in local regions rather than optimal representation can be used to segment defects in textured materials. These optimal filters cannot explicitly detect the defects but makes the detection an easier task by greatly attenuating pixel value in defect free region relative to regions having defects. The closed-form solution suggested by Mahalanobis-Singh and related approach by Unser has been investigated for defect detection. The approximate closed-form solution suggested by Randen and Husøy for optimal energy separation using Fisher criterion has been used. Two quantitative measures, i.e. minimum size mask and the misclassification rate, were introduced for the performance evaluation of the optimal filters designed with different object functions. One of the important conclusions of this work is that the size of optimal filter has appreciable effect on the performance for the defect detection. Some details and contributions from this work can also be found in the paper published from IEEE Trans. Systems, Man, and Cybernetics.
Figure 3: Fabric samples with big-knot, double-weft, broken-yarn and tripe-warp and their detection using optimal FIR filters.
Figure 4: Detection of mispick with just 3 ´ 3 optimal filter mask.
Figure 4: Inspection results from proposed online defect detection scheme using warp-weft optimal FIR filters.