Neural network based detection of defects in texture surfaces

In this article we present an algorithm for automatic detection of surface defects on ceramic tiles. This algorithm is based on the probabilistic neural network with radial basis. To improve sensitivity of the detection procedure an image of the tile is divided into segments and one neural network is made for each segment. The discrete wavelet transform (DWT) is used for the feature extraction in every segment. Maximums of the wavelet coefficients as well as the mean value of the approximation coefficients form an input vector for the neural network. Experimental results of the defect detection for different types of tiles and with different parameters of the algorithm show a high sensitivity and applicability of the proposed procedure.