Texture defect detection using support vector machines with adaptive Gabor wavelet features

This paper aims at investigating a method for detecting defects on textured surfaces using a Support Vector Machines (SVM) classification approach with Gabor wavelet features. Instead of using all the filters in the Gabor wavelets, an adaptive filter selection scheme is applied to reduce the computational cost on feature extraction while keeping a reasonable detection rate. One-Against-All strategy is adopted to prepare the training data for a binary SVM classifier that is learnt to classify pixels as defective or non-defective. Experimental results on comparison with other multiresolution features and the Learning Vector Quantization (LVQ) classifier demonstrate the effectiveness of the proposed method on defect detection on textured surfaces.