The Algorithm of ICA Based on PCA for Fabric Defect Detection

Abstract

The Independent Component Analysis (ICA) algorithm based on Principal Component Analysis (PCA)\r is described in this paper to achieve the raw textile defect detection. In the first step, the observed\r matrix X is constructed from a large number of defect-free sub-images and PCA is operated to achieve\r dimension reduction. In the second step, the transformation matrix W and independent basis subspace\r s are obtained from defect-free sub-images through ICA. In the final step, feature extraction is achieved\r from the overlapping sub-windows of a test image. Then a sub-window is classified as defective or nondefective\r according to Euclidean distance. The results have been analyzed in detail and illustrated this\r approach has better performance in raw textile.