An Feed Forward Model for Fabric Defect Detection using Region based Texture Similarity Matrix and Gabor Filters

Lot of approaches has been discussed for the detection of fabric defects but suffers with the problem of detection accuracy, because most of them uses texture based methods which suffers to identify the presence of micro defects. To overcome the issue of fabric defect detection, an feed forward model has been discussed in this paper, which feeds the learned feature to the next layer. At first stage, the input image is applied with Gabor filters to remove the noise introduced by capturing device or any form of communication. Then, the proposed method construct iterative sub sampling image from the noise removed image. At each level of sub sampling, the method uses different size to generate sub sampling image and from the generated sub sampling image, the method constructs region based texture similarity matrix. The number of level being used is depending on the accuracy needed by the classification approach. Using generated multi region based texture matrix, the method computes similarity with the input sub sampled image. Based on the deviation present in the input image at different level texture matrix, the method identifies the defect present in the input fabric image. The method increases the performance of fabric defect detection by improving the accuracy and reduces the false classification ratio.