TY - JOUR
AU - Renuka Devi,, M
AB - Abstract A fabric fault detection scheme is implemented in this work using frame harmonizing-based approach. Many recent research works deliver the fabric defect detection techniques. But in the textiles industry, the fast process of fabric cutting and sewing, lots of auxiliary imperfections have emerged. Particularly these defects cannot be identified easily by experts as well as an automated system. In our system, a novel frame extraction technique is used to find the defects in the fabric production pipeline. The input fabric image frame is pre-processed by transformation and filtering techniques. In this work,a novel multi-directional and multi-scale outline extraction method is proposed to extract the edge map. Contour-based features are extracted and classified by K-nearest neighbor classifier. The experimentation with real-time dataset produced the outstanding performance results when compared with state of the art methods. 1. INTRODUCTION The ceaselessly changing style of garments has created a more noteworthy product variety and shorter life cycle for a generation. Fabric defect detection is a critical process in the quality control section. Fabric inspection by human sight is easily influenced by the physical quality and mental status of the auditors. Under the wild rivalry in the textile industry, the quality assurance is an essential one to promote competitive advantage of the product. In textile quality sector, automated fabric defect detection has turned out to be more interesting domain. Numerous methodologies were proposed to address this issue, detecting defects on the fabrics. The improvement of an adaptable, expert, reliable, combined and continuous vision framework for quality control process for textile manufacturers is a basic issue. There are various model-based strategies that are available for defect detection on the fabric. We have reviewed the research works related to, on the fabric, defect detection process. The linear wave scanning mechanism was used to monitor the transmitted waves with Piezo ceramic sensors [1] in fabric inspection system. Blob detection technique [2] is used as a representative of a structural texture analysis approach and it accounts for the aspect of human visual system in detecting large varieties of textile flaws. Defects are detected by observing a threshold of acceptable differences in the properties of blobs based on human perception. An automatic analysis of woven fabric structure and fabric quality is evaluated by automatic fabric evaluation system [3]. Local contrast deviation approach [4] based on the features of fabric structure for fabric defect detection was proposed to describe features of the contrast difference in four directions between the analyzed image and a defect-free image of the same fabric and is used with a bi-level threshold function for defect segmentation. To remove the fabric background and isolate the defects, Gabor wavelet [5] based texture features are used. A system has been trained by these texture features to construct the structuring elements. The texture features are the key values to detect the defects in the textured image. In the work [6], first the co-occurrence features are extracted from the decomposed fabric-textured image and then a Mahalanobis distance classifier is trained based on the features of annotated samples. A strategy for non-destructively assessing the integrity [7] of structures is utilized for the estimation of basic normal frequencies. Another new data fusion [8] plan to multiplex the data from the distinctive channels was proposed by communicating the exchange between execution and computational load. In view of the enhanced binary, textural and neural network algorithms [9] were yielded great outcomes in the discovery of numerous sorts of fabric defects under genuine modern conditions, where the closeness of numerous kinds of fabric faults is inescapable wonder. A simulated fabric model [10] was utilized to comprehend the connection between the texture structure in the picture space and in the recurrence space in which Fourier transforms are connected to monitor the spatial frequency spectrum of a fabric. The programmed investigation of imperfections in haphazardly textured surfaces [11] local highlights of textures it depends on a specific global image reconstruction scheme using the Fourier transform. The authors of [12] presented a multi determination approach for the assessment of homogeneously textured surfaces based on an efficient image restoration scheme using the wavelet transform. In developing a new defect detection scheme [13], the Gabor filters are planned based on the texture features extracted ideally from a non-imperfect fabric image by using a Gabor wavelet network. The execution of their deformity detection scheme evaluated off-line by utilizing a set of fabric images taken from a database comprising of a wide assortment of fabric images. A PC-based real-time [14] inspection framework was proposed with advantages of minimal effort and high-detection rate. A framework for the automated visual investigation of textiles and a fault detection algorithm was proposed in [15, 16]. Modern vision frameworks must work progressively and adaptable to suit varieties in examination destinations. This was the basis for developing a detection algorithm that utilizes frame harmonizing strategy as key for structural defects of fabric. As a rule, what a computer vision appraisal framework investigates and controls is that of particular objects. Along these lines, we can pick satisfactory innovation and rearrange the execution procedure based on the earlier learning of the performed basic items, which is likewise adequately comprehended to accomplish the objective of the programmed task. As per the earlier information separated from the shape of fabric frame, we utilize outline orchestrating construct approach based on texture imperfection review as follows. The methodology can be divided into five sections: In Section 2.2, image pre-processing techniques used for enhancement are explained. Section 2.3 describes our proposed multi-directional and multi-scale (MDMS) edge detector to detect edge map. In Section 2.4, extraction of contour features is described. Section 2.5 provides the classification technique applied for best results. 2. METHODOLOGY Therefore, in this study, we have developed a system for automatic fabric part structural defect detection using a novel edge detection method. For this purpose, real-time fabric images are gathered from the fabric manufacturing industries such as ‘Hershees Men’s Designer’ and ‘Finela Designers’. To rectify the distortions while capturing the input image, the affine transformation is used. To enhance the edges, guided filter is used as pre-processing technique. MDMS is used for finding the boundaries of fabric in the pre-processed images. The edge map of an image carries the solid structure of the image, independent of the color attribute. Based on the statistical and transformation-based contour features are extracted from the edge map then KNN classifier is trained for classifying the defect category. Figure 1 shows the defect detection architecture. FIGURE 1. View largeDownload slide System architecture. FIGURE 1. View largeDownload slide System architecture. 2.1. Dataset description The real-time fabric images are gathered from the fabric manufacturing industries such as ‘Hershees Men’s Designer’ and ‘Finela Designers’. According to the statement of fabric production executives and administrators, the fabric defects are mainly identified while on the fabric but it is too difficult to detect the structural defects. So the fabric parts of each fabric are first listed out and subsequently collected 10 samples of each fabric parts. 15 fabric types, which comprises of 64 fabric parts in total. So the 640 images are collected. The images are captured using the standard 7 megapixel (Brand&Model:Canon A620) digital camera with fixed one-meter distance and angle of 90 degrees to the fabric part surface. The resolution of an image is 3072 × 2304. The entire dataset is labeled as per the state of the fabric whether it is defected or not. The standard fabric part also captured for all 64 parts. So the fabric database totally has 704 images. 2.2. Pre-processing 2.2.1. Affine transformation Before involved with the edge detection process, the pre-processing steps are executed to make the input image invariant with size and rotation. Although the camera and fabric input frame are fixed, the geometric distortions may occur with non-ideal placement of fabric on the frame. To avoid such distortions, in our work, an affine transformation is applied. Affine transformation is a linear mapping method that processes the input fabric image and produces the standard invariant input image. As stated earlier, even though, the capturing camera and input frame are maintained in an ideal location there may be a chance for shearing or translation effect in the fabric on the input frame. So, affine transformation [17] is applied to the input image frame. The resultant transformed image will be used by further steps. 2.2.2. Guided filter To enrich the edge detection process border pixels of the fabric parts have to be enhanced. The linear translation-variant guided filtering [18] process makes this enhancement perfectly. The guidance image I and filtering input image K are the parameters to the filtering process. In our work, both I and K are identical based on the function specified earlier in the reference [18]. The filtering output at a pixel i is articulated as a weighted average: Mi=∑jWij(I)Kj (1) where i and j represent the pixel indices. M is the filtered image filtered with the filtering kernel Wij ⁠, which is linear with respect to K. 2.3. MDMS edge detector This novel edge detection scheme is evolved from the classical Berkeley edge detector [19]. It applied to intensity or color images by combining color, brightness and texture cues that provide a probabilistic edge map. For each pixel in the image, a probability of being an edge or contour is calculated. The algorithm works by looking for local discontinuities in several feature channels (brightness and texture) and over a range of orientations and scales, at each image pixel. However, the probability is being estimated based on the circle with radius r and divides it along the diameter and orientation. The G(x,y,θ,r) is the gradient function, which compares the content of the two resulting disc halves. A large difference between the disc halves indicates a discontinuity in the image along the disc’s diameter [20]. The content of the disk halves is represented as histogram by the three pixel-based features such as brightness, color and texture. Whereas our proposed MDMS edge detection scheme operates in multi-direction and multi-scale manner. Let G(x,y,θ,r) be the gradient function operates based on the orientation θ and circle radius r. The new derived MDMS scheme predicts the probability based on the gradient function G(x,y,φ,ϕ) where ϕ is set of all orientations involved in gradient function and φ is set of all radius involved in the gradient function. Then the difference operator X2 is used to compare the histogram and halves and is shown in Equation (2). X2(g,h)=∑a∈φ∑b∈ϕ∑(giab−hiab)2/giab+hiab2 (2) As mentioned previously, the three features are selected. The detector uses the CIELAB (CIE L*a*b*) feature space [20] defined by the International Commission on Illumination (CIE). Hence, for representing brightness the L* (luminance) is chosen and for color representation, the chrominance components a* and b* are used. For better accuracy, while maintaining computational cost low, the color gradient is calculated as the sum of the gradients in the a* and b* axes rather than calculating the joint histogram. For the texture gradient, histograms of vector quantized filter outputs are calculated and compared. Multiple oriented and scaled features are extracted and learned as logistic regression model to determine the final edge probability for each pixel. With an activated threshold, the edge pixels are retained for further feature extraction process. 2.4. Contour-based feature extraction To learn the structural properties of a fabric frame, the contour-based features are important to extract. Among the numerous impressive features in contour, few eminent descriptors are taken into account. Totally, we have examined 12 kinds of descriptors [21] such as minimum bounding rectangle, circle variance, circularity ratio, convexity, rectangularity, cumulative angular deviant, contour curvature, ellipse variance, average bending energy, boundary moments, beam angle statistics (BAS) and chord distribution. These feature vectors are used to detect the defects in fabric parts. 2.4.1. Minimum bounding rectangle The minimum bounding rectangle is also called minimum bounding box. It is the smallest rectangle that contains every point in the extracted fabric frame. For an arbitrary shape, eccentricity is the ratio of the length L and width W of the minimal bounding rectangle of the shape at some set of orientations. Elongation, Elo ⁠, is another concept based on eccentricity. Elongation is a measure that takes values in the range [0; 1]. Asymmetrical shape in all axes such as a circle or square will have an elongation value of 0 whereas shapes with large aspect ratios will have an elongation closer to 1. Elo=1−W/L (3) 2.4.2. Circularity ratio Circularity ratio represents how a fabric shape is similar to a circle. There are two definitions: circularity ratio is the ratio of the area of the fabric frame to the area of a circle having the same perimeter: C1=AsAc (4) where As is the area of the extracted frame and Ac is the area of the circle having the same perimeter as the frame. Assume the perimeter is O, so Ac=O2/4π ⁠. Then C1=4π⋅As/O2 ⁠. As 4π is a constant, so we have the second circularity ratio definition. Circularity ratio is the ratio of the area of a shape to the shape’s perimeter square: C2=AsO2 (5) 2.4.3. Circle variance The circle variance of the fabric frame is defined as: Cva=σRμR (6) where σR and μR are the mean and standard deviation of the radial distance from the centroid (gx,gy) of the frame to the boundary points (xi,yi) ⁠; i∈[0,N−1] ⁠. They are the following formulae, respectively: μR=1N∑i=1N−1di (7) σR=1N∑i=1N−1(di−μR)2 (8) where di=(xi−gx)2+(yi−gy)2 ⁠. So the most compact shape is a circle. 2.4.4. Convexity Convexity is defined as the ratio of perimeters of the convex hull Oconvexhull over that of the original fabric frame contour O convexity=OconvexhullO .(9) The region R2 is convex if and only if for any two points P1P2∈R2 the whole line segment P1P2 is inside the region. The convex hull of a region is the smallest convex region including it. 2.4.5. Rectangularity Rectangularity (R) represents how rectangular the fabric frame is, i.e. how much it fills its minimum bounding rectangle: R=AsAR (10) where As is the area of a fabric frame; AR is the area of the minimum bounding rectangle of fabric frame As ⁠. 2.4.6. Cumulative angular deviant A periodic function is termed as the cumulative angular deviant function ψ(t) and is defined as ψ(t)=ϕ(N2πt)−tt∈[0,2π] (11) where N is the total number of contour points. 2.4.7. Contour curvature The curvature is a very important boundary feature for anyone to judge similarity between shapes. It also has salient perceptual characteristics and has proven to be very useful for shape recognition. In order to use K(n) for shape representation, we quote the function of curvature, K(n) ⁠, from [22, 23] as: K(n)=ẋ(n)ÿ(n)−ẏ(n)ẍ(n)(ẋ(n)2+ẏ(n)2)3/2 (12) Therefore, it is possible to compute the curvature of a planar curve from its parametric representation. If n is the normalized arc-length parameter s, the above equation can be written as: K(s)=ẋ(s)ÿ(s)−ẏ(s)ẍ(s) (13) As given in the above equation, the curvature function is computed only from parametric derivatives, and, therefore, it is invariant under rotations and translations. However, the curvature measure is scale dependent, i.e. inversely proportional to the scale. By normalizing this measure by the mean absolute curvature, we can archive scale independence, i.e. K′(s)=K(s)1N∑s=1N|K(s)| (14) where N is the number of points on the normalized contour. When the size of the curve is an important discriminative feature, the curvature should be used without the normalization; otherwise, for the purpose of scale-invariant shape analysis, the normalization should be performed. An approximate arc-length parameterization based on the centripetal method is given by the following [23]: Let P=∑n=1Ndn be the perimeter of the curve and L=∑n=1Ndn ⁠, where dn is the length of the chord between points Pn and Pn+1,n=1,2,…,N−1 ⁠. The approximate arc-length parameterization relations: S1 = 0 sk=sk−1+Pdk−1L,k=2,3,…,N (15) Starting from an arbitrary point and following the contour clockwise, we compute the curvature at each interpolated point using Equation (15). Evidently, as a descriptor, the curvature function can distinguish different shapes. 2.4.8. Ellipse variance Ellipse variance Eva is a mapping error of a shape to fit an ellipse that has an equal covariance matrix as the shape [21]: C=1N∑i=0N−1(xi−gxyi−gy)(xi−gxyi−gy)T=(cxx−cxycyx−cyy) (16) where cxy=1N∑i=0N−1(xi−gx)(yi−gy)cxx=1N∑i=0N−1(xi−gx)2,cyx=1N∑i=0N−1(yi−gy)(xi−gx) cyy=1N∑i=0N−1(yi−gy)2 and Cellipse=C ⁠. It is practically effective to apply the inverse approach yielding. We assume Vi=(xi−gxyi−gy) ⁠, di′=ViT⋅Cellipse−1⋅Vi ⁠, μR′=1N∑i=1N−1di′ and σR′=1N∑i=1N−1(di′−μR′)2 Then Eva=σR′μR′ (17) Comparing with circularity ratio, Ellipse variance Eva represents a fabric frame more accurately than circularity ratio. 2.4.9. Average bending energy Assume K(s) is the curvature function, S is the arc-length parameter and N is the number of points on a contour, [24] the average bending energy BE is defined by BE=1N∑s=0N−1K(s)2 (18) In order to compute the average bending energy more efficiently, Young et al. [25] did the Fourier transform of the boundary and used Fourier coefficients and Parseval’s relation. 2.4.10. Boundary moments The advantage of boundary moment descriptors is that it is easy to implement. However, it is difficult to associate higher order moments with the physical interpretation of the fabric frame. The boundary moments, analysis of a contour, can be used to reduce the dimension of boundary representation [26]. Assume shape boundary has been represented as a 1D shape representation z(i) as introduced, the rth moment mr and central moment μr can be estimated as mr=1N∑i=1N[z(i)]r (19) μr=1N∑i=1N[z(i)−m1]r (20) where N is the number of boundary points. The normalized moments m¯r=mr/(μ2)r/2 and μ¯r=μr/(μ2)r/2 are invariant to shape translation, rotation and scaling. Less noise-sensitive shape descriptors of the fabric frame can be obtained from F1=(μ2)r/2m1,F2=μ3(μ2)3/2,F3=μ4(μ2)2. (21) 2.4.11. Beam angle statistics BAS shape descriptor is based on the beams originated from a fabric boundary point [27]. BAS shape descriptor captures the perceptual information using the statistical information based on the beams of individual points. It gives globally discriminative features to each boundary point by using all other boundary points. BAS descriptor is also quite stable under distortions and is invariant to translation, rotation and scaling. Let B be the fabric boundary. B={P1,P2,…,PN,} is represented by a connected sequence of points, P1=(xi,yi) ⁠, i=1,2,…,N ⁠, where N is the number of boundary points. For each point Pi ⁠, the beam angle between the forward beam vector vi+k=pipi+k→ and backward beam vector vi−k=pipi−k→ in the kth order neighbor system is then computed as ck(i)=(θvi+k−θvi−k) (22) where θvi+k=arctanyi+k−yixi+k−xi and θvi−k=arctanyi−k−yixi−k−xi ⁠. For each boundary point Pi of the contour, the beam angle ck(i) can be taken as a random variable with the probability density function p(ck(i)) ⁠. Therefore, the fabric frame compact representation can be provided by BAS. 2.4.12. Chord distribution By calculating the lengths of all chords in the shape, that is all pairwise distances between boundary points, the chord distribution is generated as a ‘lengths’ histogram, which is invariant to rotation and scales linearly with the size of the object [28]. The highest 10 bin values are taken as the features descriptors, which discriminate the fabric frame in accurate manner. 2.5. Defect detection process After extraction of the feature vector from the fabric frame, these features are used by the defect detection process framework. For this purpose, K-nearest neighbor (KNN) classification algorithm is used. KNN is an instance-based learning classifier that performs classification based on the closest data point in feature space. Consider the X: training data, Y: class labels of X, x: unknown samples. X and Y consist of m samples then D is the vector of distance between X and x. The smallest k distance sample’s labels are computed as set S, so S is the subset of Y. The majority label y is identified the class, where y∈S ⁠. In our system, the extracted features are fed into the fitted KNN classifier for detecting the fabric defect. We have implemented and compared few other supervised classifiers such as support vector machine (SVM), restricted Boltzmann machine (RBM) and extreme learning machine (ELM). 3. EXPERIMENTAL RESULTS This section describes the experimental setup and provides the results and analysis of the proposed approach for classification problem of fabric defect detection. The experiments were conducted on the collected fabric image dataset. The images were studied under the contour frame harmonizing procedures. Using the contour-based features, the KNN classifier has been trained and tested for the performance evolution. The proposed system is implemented using the MATLAB software. The training and testing are carried out in the Windows 8.1 PC configured by Intel Core I3 CPU 2.3 GHz, 2 GB RAM and 250 GB storage. In this work, we have focused on the classification of fabric defects. The classification process classifies the fabrics parts as normal or defected. A new method MDMS is used to extract the edge map of the fabric structure. The dataset is consisting of standard parts of fabric types that are reliably used by the training process based on the extracted feature descriptors. The defected structural properties of the fabric parts are also learned by the classifier using the defected samples. A total of 64 types of fabric parts were involved to train the system comprising of 320 standards (not defected) and 320 defected images. The test set comprising 420 (230 normal and 190 defected) images. The test sets are evolved based on the random sampling method. This section gives the results for the classification using KNN classifier. The confusion matrices are given for the classification of training and test data. In Table 1, we provide the confusion matrices obtained during the testing phase. From the confusion matrices, it is clear that the proposed approach yields good results for test sets. The classification results in the test set are precise. It is also clear from the matrices that although the proposed approach leads to higher accuracies for the classification of fabric defect. The observation leads to the fact that, the, not defected fabric frames are misclassified as defected because of more curvatures present in the fabric frame. The same scenario is also observed in the false positive cases. TABLE 1. Confusion matrices obtained to evaluate the system while testing. Classified type Normal Defected Test set  Normal 221 9  Defected 10 180 Classified type Normal Defected Test set  Normal 221 9  Defected 10 180 View Large TABLE 1. Confusion matrices obtained to evaluate the system while testing. Classified type Normal Defected Test set  Normal 221 9  Defected 10 180 Classified type Normal Defected Test set  Normal 221 9  Defected 10 180 View Large The statistical performance measures such as sensitivity, specificity, F-score, precision and accuracy are measured to evaluate the proposed architecture. We design the comparison between our proposed architecture and three other methods, i.e. SVM, RBM and ELM. The analysis of performance measures among proposed MDMS and Berkeley edge detector with different classifiers are listed in Table 2. Figure 2 illustrated the classification performance of our proposed method along with other compared classifiers. In these evaluations, the same feature descriptors are used for different analysis. The experimental results of the proposed method attains 96.09% of sensitivity, 94.74% of specificity, 95.67% of precision, 95.48% of accuracy and 95.88% of F-score. TABLE 2. Performance measures of the proposed work and other classifiers for the test set. Classifier Proposed MDMS edge detection Berkeley edge detector [19] Sensitivity Specificity Precision Accuracy F-score Sensitivity Specificity Precision Accuracy F-score Proposed (KNN) 0.9609 0.9474 0.9567 0.9548 0.9588 0.9521 0.9295 0.9421 0.9484 0.9391 SVM 0.9589 0.9314 0.9487 0.9487 0.9503 0.9452 0.9128 0.9354 0.9214 0.9217 RBM 0.9452 0.9305 0.9329 0.9321 0.9267 0.9254 0.9183 0.9301 0.9195 0.9142 ELM 0.9245 0.9123 0.9012 0.9142 0.9034 0.9195 0.9114 0.9011 0.9011 0.8912 Classifier Proposed MDMS edge detection Berkeley edge detector [19] Sensitivity Specificity Precision Accuracy F-score Sensitivity Specificity Precision Accuracy F-score Proposed (KNN) 0.9609 0.9474 0.9567 0.9548 0.9588 0.9521 0.9295 0.9421 0.9484 0.9391 SVM 0.9589 0.9314 0.9487 0.9487 0.9503 0.9452 0.9128 0.9354 0.9214 0.9217 RBM 0.9452 0.9305 0.9329 0.9321 0.9267 0.9254 0.9183 0.9301 0.9195 0.9142 ELM 0.9245 0.9123 0.9012 0.9142 0.9034 0.9195 0.9114 0.9011 0.9011 0.8912 View Large TABLE 2. Performance measures of the proposed work and other classifiers for the test set. Classifier Proposed MDMS edge detection Berkeley edge detector [19] Sensitivity Specificity Precision Accuracy F-score Sensitivity Specificity Precision Accuracy F-score Proposed (KNN) 0.9609 0.9474 0.9567 0.9548 0.9588 0.9521 0.9295 0.9421 0.9484 0.9391 SVM 0.9589 0.9314 0.9487 0.9487 0.9503 0.9452 0.9128 0.9354 0.9214 0.9217 RBM 0.9452 0.9305 0.9329 0.9321 0.9267 0.9254 0.9183 0.9301 0.9195 0.9142 ELM 0.9245 0.9123 0.9012 0.9142 0.9034 0.9195 0.9114 0.9011 0.9011 0.8912 Classifier Proposed MDMS edge detection Berkeley edge detector [19] Sensitivity Specificity Precision Accuracy F-score Sensitivity Specificity Precision Accuracy F-score Proposed (KNN) 0.9609 0.9474 0.9567 0.9548 0.9588 0.9521 0.9295 0.9421 0.9484 0.9391 SVM 0.9589 0.9314 0.9487 0.9487 0.9503 0.9452 0.9128 0.9354 0.9214 0.9217 RBM 0.9452 0.9305 0.9329 0.9321 0.9267 0.9254 0.9183 0.9301 0.9195 0.9142 ELM 0.9245 0.9123 0.9012 0.9142 0.9034 0.9195 0.9114 0.9011 0.9011 0.8912 View Large FIGURE 2. View largeDownload slide Illustrated the classification performance of our proposed method along with other compared classifiers. FIGURE 2. View largeDownload slide Illustrated the classification performance of our proposed method along with other compared classifiers. 4. CONCLUSION In this work, computer-aided image analysis for fabric part structural defect classification is proposed. This analysis makes use of computational methods that employ various image processing techniques. The input fabric image frame is pre-processed by transformation and filtering techniques. Our novel outlines extraction method MDMS is proposed to extract the edge map. Contour-based features are extracted and classified by KNN classifier. The experimentation with collected dataset produced the better performance results when compared with other classifiers. The specific outcome of this research is an inspection or detection system that detects the structural defects of the fabric parts. Thus the system can be used as an inspection system in any fabric production industries. In future, we have planned to implement the frame matching techniques to detect the defect in the fabric frame. REFERENCES 1 Kessler , S.S. , Spearing , S.M. and Soutis , C. ( 2002 ) Damage detection in composite materials using Lamb wave methods . Smart Mater. Struct. , 11 ( 2 ), 269 . Google Scholar Crossref Search ADS WorldCat 2 Bodnarova , A. , Bennamoun , M. and Kubik , K.K. ( 1998 ) ‘Defect Detection in Textile Materials Based on Aspects of the HVS.’ In IEEE Int. Conf. Systems, Man, and Cybernetics, 1998, Vol. 5, pp. 4423–4428. 3 Kang , T.J. , Choi , S.H. , Kim , S.M. and Oh , K.W. ( 2001 ) Automatic structure analysis and objective evaluation of woven fabric using image analysis . Text. Res. J. , 71 , 261 – 270 . Google Scholar Crossref Search ADS WorldCat 4 Shi , M. , Fu , R. , Guo , Y. , Bai , S. and Xu , B. ( 2011 ) Fabric defect detection using local contrast deviations . Multimed. Tools Appl. , 52 ( 1 ), 147 . Google Scholar Crossref Search ADS WorldCat 5 Mak , K.-L. , Peng , P. and Yiu , K.F.C. ( 2009 ) Fabric defect detection using morphological filters . Image Vis. Comput. , 27 , 1585 – 1592 . Google Scholar Crossref Search ADS WorldCat 6 Cawley , P. and Adams , R.D. ( 1979 ) The location of defects in structures from measurements of natural frequencies . J. Strain Anal. Eng. Des. , 14 , 49 – 57 . Google Scholar Crossref Search ADS WorldCat 7 Tandon , N. and Choudhury , A. ( 1999 ) A review of vibration and acoustic measurement methods for the detection of defects in rolling element bearings . Tribol. Int. , 32 , 469 – 480 . Google Scholar Crossref Search ADS WorldCat 8 Mahajan , P.M. , Kolhe , S.R. and Patil , P.M. ( 2009 ) A review of automatic fabric defect detection techniques . Adv. Comput. Res. , 1 , 18 – 29 . WorldCat 9 Kumar , A. and Pang , G.K.H. ( 2002 ) Defect detection in textured materials using Gabor filters . IEEE Trans. Ind. Appl. , 38 , 425 – 440 . Google Scholar Crossref Search ADS WorldCat 10 Chan , C.-H. and Pang , G. ( 1999 ) ‘Fabric Defect Detection by Fourier Analysis.’ In Industry Applications Conf., 1999. Thirty-Fourth IAS Annual Meeting. Conference Record of the 1999 IEEE, Vol. 3, pp. 1743–1750. 11 Tsai , D.-M. and Huang , T.-Y. ( 2003 ) Automated surface inspection for statistical textures . Image Vis. Comput. , 21 , 307 – 323 . Google Scholar Crossref Search ADS WorldCat 12 Tsai , D.-M. and Chiang , C.-H. ( 2003 ) Automatic band selection for wavelet reconstruction in the application of defect detection . Image Vis. Comput. , 21 , 413 – 431 . Google Scholar Crossref Search ADS WorldCat 13 Mak , K.L. and Peng , P. ( 2008 ) An automated inspection system for textile fabrics based on Gabor filters . Rob. Comput. Integr. Manuf. , 24 , 359 – 369 . Google Scholar Crossref Search ADS WorldCat 14 Cho , C.-S. , Chung , B.-M. and Park , M.-J. ( 2005 ) ‘Development of real-time vision-based fabric inspection system.’ . IEEE Trans. Ind. Electron. , 52 , 1073 – 1079 . Google Scholar Crossref Search ADS WorldCat 15 Jing , J. , Zhang , H. , Wang , J. , Li , P. and Jia , J. ( 2013 ) Fabric defect detection using Gabor filters and defect classification based on LBP and Tamura method . J. Text. Inst. , 104 , 18 – 27 . Google Scholar Crossref Search ADS WorldCat 16 Abouelela , A. , Abbas , H.M. , Eldeeb , H. , Wahdan , A.A. and Nassar. , S.M. ( 2005 ) Automated vision system for localizing structural defects in textile fabrics . Pattern Recognit. Lett. , 26 , 1435 – 1443 . Google Scholar Crossref Search ADS WorldCat 17 Zhang , G.M. , Xu , J.Y. and Liu , J.X. ( 2015 ) A New Method for Recognition Partially Occluded Curved Objects Under Affine Transformation. In 10th Int. Conf. Intelligent Systems and Knowledge Engineering (ISKE), 2015, pp. 456–461. 18 He , K , Sun , J and Tang , X ( 2013 ) Guid22ed image filtering . IEEE Trans. Pattern Anal. Mach. Intell. , 35 , 1397 – 1409 . Google Scholar Crossref Search ADS PubMed WorldCat 19 Martin , D.R. , Fowlkes , C.C. and Malik , J. ( 2004 ) Learning to detect natural image boundaries using local brightness, color, and texture cues . IEEE Trans. Pattern Anal. Mach. Intell. , 26 , 530 – 549 . Google Scholar Crossref Search ADS PubMed WorldCat 20 Puzicha , J. , Buhmann , J.M. , Rubner , Y. and Tomasi , C. , 1999 . Empirical Evaluation of Dissimilarity Measures for Color and Texture. Proc. Seventh IEEE Int. Conf. Computer Vision, Vol. 2, pp. 1165–1172. 21 Yang , M. , Kpalma , K. and Ronsin , J. ( 2008 ). A survey of shape feature extraction techniques . Pattern Recognit. IN-TECH, 43 – 90 . https://hal.archivesouvertes.fr/hal-00446037 22 Mokhtarian , F. and Mackworth , A.K. ( 1992 ) A theory of multiscale, curvature-based shape representation for planar curves . IEEE Trans. Pattern Anal. Mach. Intell. , 14 , 789 – 805 . Google Scholar Crossref Search ADS WorldCat 23 Jalba , A.C. , Wilkinson , M.H.F. and Roerdink , J.B.T.M. ( 2006 ) Shape representation and recognition through morphological curvature scale spaces . IEEE Trans. Image Process. , 15 , 331 – 341 . Google Scholar Crossref Search ADS PubMed WorldCat 24 Loncaric , S. ( 1998 ) A survey of shape analysis techniques . Pattern Recognit. , 31 , 983 – 1001 . Google Scholar Crossref Search ADS WorldCat 25 Young , I.T. , Walker , J.E. and Bowie , J.E. ( 1974 ) An analysis technique for biological shape I . Inf. Control , 25 , 357 – 370 . Google Scholar Crossref Search ADS WorldCat 26 Sonka , M. , Hlavac , V. and Boyle , R. ( 2014 ) Image processing, analysis, and machine vision’ . Cengage Learning . Google Preview WorldCat COPAC 27 Arica , N. and Vural , F.T.Y. ( 2003 ) BAS: a perceptual shape descriptor based on the beam angle statistics . Pattern Recognit. Lett. , 24 , 1627 – 1639 . Google Scholar Crossref Search ADS WorldCat 28 Smith , S.P. and Jain , A.K. ( 1982 ) Chord distributions for shape matching . Comput. Graph. Image Process. , 20 , 259 – 271 . Google Scholar Crossref Search ADS WorldCat © The British Computer Society 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Detection of Structural Defects in Fabric Parts Using a Novel Edge Detection Method
JF - The Computer Journal
DO - 10.1093/comjnl/bxy121
DA - 2019-07-18
UR - https://www.deepdyve.com/lp/oxford-university-press/detection-of-structural-defects-in-fabric-parts-using-a-novel-edge-PI0FPLVxwH
SP - 1036
VL - 62
IS - 7
DP - DeepDyve
ER -