On the Use of Discrete Wavelets in Implementing Defect Detection System for Texture Images

In this paper, a novel wavelet-based approach to the detection of defects in grey-level texture images is proposed. This new approach (system) explores specific properties of the discrete wavelet transform (DWT), evaluates the statistical analysis results associated with well-defined and task-oriented subsets of DWT spectral coefficients, and generates defect detection criteria which, in their turn, evaluate many-sided nature of potential defects in texture images and leave space for controlling the risk, i.e. for controlling the percentage of false positives and/or false negatives in a particular class of texture images. The experimental results demonstrating the use of the proposed system for the visual inspection of ceramic tiles, obtained from the real factory environment, and textile fabric scraps are also presented.