An adaptive level-selecting wavelet transform for texture defect detection

We present an effective approach based on wavelet transform (WT) to detect defects on images with high frequency texture background. The original image is decomposed at various levels by WT. Then, by selecting an appropriate level at which the approximation sub-image is reconstructed, textures on the background are effectively removed. Thus, the difficult texture defect detection problem can be settled by non-texture techniques. An adaptive level-selecting scheme is presented by analyzing the co-occurrence matrices (COM) of the approximation sub-images. Experiments are done to detect the stains and broken points on texture surfaces. Comparisons with frequency domain low and high pass filters show that our method is much more effective.