Wavelet methods for texture defect detection

In this article we introduce our approach to exploit multiscale wavelet methods for texture defect detection. Several wavelet bases and decomposition algorithms are examined in regard of applicability, parameterization and computational costs. The article points out specific problems in localizing texture defects in multiscale wavelet representations. Besides the fast dyadic wavelet transform we demonstrate the application of the translation invariant a trous algorithm on texture samples. Feature extraction methods are proposed and examples of successful defect classification results are shown.