A Novel Approach for Detecting Defects of Random Textured Tiles Using Gabor Wavelet

In this paper we address the problem of detecting different type of defects on random textured tiles. We adopt Gabor wavelet for analysis of random textured surfaces. Unlike the existing methods which arrange the Gabor filter Bank in such a way that the half-magnitude contour of neighboring filters in frequency domain touch each other, we allows the bandwidth of filters to vary. This flexibility enables the designed filter bank to detect different types of defects. We demonstrate that the proposed method detects defects on random textured tiles better than the conventional Gabor wavelet method. Key word: Texture analysis Defect Detection Random texture tiles Gabor wavelet INTRODUCTION on twelve samples of color random texture tiles was Automated inspection of production line has been Hocenski et al. [6]. received many attentions nowadays. Among the existing A defect detection method based on texture analysis applications automatic inspection of ceramic tiles, textile was proposed for random texture tiles by Xie et al [7]. and porcelain are very important because of significant In this work it is assumed that each image is generated by economic and labor saving benefits. Moreover the a superposition of various-size image patches. Novelty customers constantly require higher quality products and detection on color texture surfaces is performed by human defect detection is subject to error for many examining the same-source similarity based on the data reasons: relatively high speed of conveying belt and likelihood in multi-scale, followed by logical processes to harsh environment for workers etc. So, there are large combine the defect candidates to localize defects. The influences of human errors and subjectivity on the result method needs to learn normal texture on tiles using a of inspection [1]. Presence of other factors such as noise, number of samples in advance [7]. Novak et al. [8] non-uniform illumination and variety of defect types in tile proposed a method based on statistical features to detect make the defect detection a challenging problem [2]. This defects on random texture tiles. The authors extract a difficulty becomes even more when tiles are random texture feature vector using Local Binary Pattern (LBP) textured. operator for defect detection. A set of ten normal and ten This paper addresses the problem of defect detection defected tiles have been used for the training of proposed for random texture tiles. Different solutions with different system. The type of defects which the proposed method levels of complexity are reported in literature. Muller and can deal with is not reported [8]. S. Rimac et al. [9] used Nickoly [3] developed a technique for the defect detection DWT in a radial neural network for defect detection in tile. using morphological processing but only for plain, This method can identify a larger group of defect but it’s non-textured surfaces. Williams et al. [4] proposed an not suitable for random texture patterns. inspection algorithm for detecting a specific type of For regular pattern textures which exhibit a high defect, their method is based on the analysis of the degree of periodicity Fourier-domain features also have image intensity histogram. In [5] a processing method been used for detection of defects [10]. However Fourier based on the singular value decomposition is reported. analysis is not suitable for detection of local defects. Vasilica et al. [1] present a method for detecting defects In fact for detection of local defects in a texture image on ceramic tiles, the proposed method detects defects such as image of tile or fabric, multi-resolution using a canny edge detector. The result of experiment decomposition of images across several scales are reported promising. Later the method was improved by