Abstract—In this article, a method has been offered to classify normal and defective tiles using wavelet transform and artificial neural networks. The proposed algorithm calculates max and min medians as well as the standard deviation and average of detail images obtained from wavelet filters, then comes by feature vectors and attempts to classify the given tile using a Perceptron neural network with a single hidden layer. In this study along with the proposal of using median of optimum points as the basic feature and its comparison with the rest of the statistical features in the wavelet field, the relational advantages of Haar wavelet is investigated. This method has been experimented on a number of various tile designs and in average, it has been valid for over 90% of the cases. Amongst the other advantages, high speed and low calculating load are prominent. Keywords—Defect detection, tile and ceramic quality inspection, wavelet transform, classification, neural networks, statistical features. I. INTRODUCTION ODAY thanks to advances in the machine visions and hardware, monitoring and classification process of industrial products can be performed automatically using digital high speed hardware and intelligent software. In the tile and ceramic industry it is possible that some different defects are appeared in tiles over different stages of the production line. Some defects such as deflection or error in size will be easily detectable by certain automatic systems. However, some other defects including those of color, dump, projections, pin holes and notches, and stains and cracks are largely detected by human operators that due to routine and wearing process of quality controlling by means of eyes will lack the needed precision more or less. Since this issue involves different aspects and applications of machine vision, pattern recognition, and image processing, it has recently taken a large amount of work and experiments to deal with. Some of them have taken advantage of image detection, image enhancement; color segmentation, shape analysis, feature extraction, and clustering. For example, Monadjemi et al. [3] have proposed a method based on special filters which do classification with high M.Ghazvini is with the University of Isfahan, Isfahan, 81746, Iran (corresponding author to provide phone: 98-311-7934035; fax: 98-311- 7932670; e-mail: ghazvini@ eng.ui.ac.ir). S. A. Monadjemi, N.Movahhedinia, K.Jamshidi are with the Unoversity of Isfahan, Isfahan, Iran;(e-mail: [monadjemi, naserm, jamshidi]@ eng.ui.ac.ir). accuracy and precision. It does not also need a large number of defected tiles to get result while reduces the training stage as well. In [1] reflective detection has been used to detect reflective surface defects by which a large number of surface defects can be detected on tiles. But this method is not able to detect some defects including print shapes and color spectrum. In recent decade, multi-precious techniques such as wavelet transform and Gabor[4] have been used largely for structure and texture analysis. Despite wavelet transform that uses fixed filter parameters for analyzing image in different scales, Gabor transform demands appropriate adjustments of filter parameters in different scales. References [5-13] include different methods for defect detection and classification of different textual images based on wavelet transform and texture features, using co-occurrence matrices of detail images and approximation resulted from wavelet transform and similar methods. Reference [2] used light fluctuation of the tile with aid of 2-dimensional wavelet transform to detect high contrast defects of the tiles in which with consideration to histogram diagram and definition of a threshold surface and its application on the image, defects in the tile is detected. Rimak. et al [14] proposed a method for ceramic defect detection based on wavelet transform and probable neural network with radial basis. They have segmented image of tile into at least two parts and for each segment have created a neural network. Vector of used features includes max of details coefficients in three points and average of approximation coefficient resulted from wavelet transform. The advantage of this method is the small size of its features vector while its disadvantage is using at least two neural networks for classification. In this study a viable algorithm with high precision and low calculating load was proposed to classify intact from defected tiles using wavelet transform and its combination with statistical features of resulted images and using a Perceptron neural network. Second part involves a review of wavelet transform and proposed algorithm for features deduction. The result appears in part three and finally part five concludes the article. II. WAVELET TRANSFORM In a kind of wavelet transform used in this study, the transform uses two filter sets of high pass and low pass and applies them on the given signal (image) in some layers. As the image crosses theses filters for the first time, four new images are acquired; approximation (result of the low pass filter), details in horizontal point, details in vertical point, and details in diameter. In order to calculate wavelet coefficients Defect Detection of Tiles Using 2D-Wavelet Transform and Statistical Features M.Ghazvini, S. A. Monadjemi, N. Movahhedinia, and K. Jamshidi T World Academy of Science, Engineering and Technology 25 2009 901