Detection of Defect in Textile Fabrics Using Optimal Gabor Wavelet Network and Two-Dimensional PCA

The aim of production line enhancement in any industry is to improve quality and reduce operating costs by applying various kinds of advanced technology. In order to become more competitive, many sensing, monitoring, and control approaches have been investigated in the textile industry. Automated visual inspection is one area of improvement where real cost savings can be realized over traditional inspection techniques. Manual visual inspection of textile products is expensive and error-prone because of the difficult working environment near the weaving machine. Automated visual detection of fabric defects is particularly challenging due to the large variety of fabric defects and their various degrees of vagueness and ambiguity. This work presents a hybrid application of Gabor filter and two-dimensional principal component analysis (2DPCA) for automatic defect detection of texture fabric images. An optimal filter design method for Gabor Wavelet Network (GWN) is applied to extract texture features from textile fabric images. The optimal network parameters are achieved by using Genetic Algorithm (GA) based on the non-defect fabric images. The resulting GWN can be deployed to segment and identify defect within the fabric image. By using 2DPCA, improvement of defect detection can significantly be obtained. Experimental results indicate that the applied Gabor filters efficiently provide a straight-forward and effective method for defect detection by using a small number of training images but still can generally handle fabric images with complex textile pattern background. By integrating with 2DPCA, desirable results have been simply and competently achieved with 98% of accuracy.