Research On Automatic Detection Methods Of Fabric Defects Based On Image Processing

Fabric defect detection is one of the important aspects in fabric production and evaluation system.Compared with the traditional manual method, the automatic fabric defect detection method based onimage processing technology can greatly improve the detection accuracy and efficiency. Studyingmore effective detection method to overcome some shortages of existing methods has theoretical andpractical significance. Researches on fabric defect image preprocessing, segmentation, featureextraction, classification and so on have been done in this thesis, which are described as follows:Firstly, a preprocessing method for fabric defect image based on complex contourlet transform(CCT), anisotropic diffusion and particle swarm optimization (PSO) is proposed. P_Laplace operatorand Catte_PM model are used as diffusion of low-frequency and high-frequency components obtainedby CCT, respectively. Then incomplete beta function optimized by PSO is applied to low-frequencyone for enhancement. The experimental results show that this method has great improvement in bothsubjective visual effect and objective quantitative evaluation indicators, which can suppress noise andenhance the texture details.Then a fabric defect image segmentation method based on pulse coupled neural network (PCNN)and symmetric Tsallis cross entropy is studied. The image is segmented by PCNN neural conditionaccording to the gray strength difference between fabric defect area and non-defect area andsymmetric Tsallis cross entropy is used as the image segmentation criterion. The experimental resultsshow that compared with Otsu, PCNN, the method based on PCNN and cross entropy, thesegmentation effect of this method is the best.And then, a fabric defect feature extraction method based on CCT and principal componentanalysis (PCA) is studied. PCA is applied to the low-frequency component with most energy of imageand part of high-frequency components with some defect detail information after CCT to get a lowerdimensional feature space. And components of testing sample obtained by CCT are projected onto thefeature space. Different types of fabric defects are distinguished by minimum Euclidean distance.Compared with PCA, the method combining wavelet with PCA and so on, this method extractsfeatures of the common fabric defect types effectively.Next, a fabric defect detection method based on Log_Gabor wavelet and Krawtchouk momentinvariants is given. A fabric defect image is processed by Log_Gabor wavelet to extract texturefeature. The vectors combined texture feature and shape feature obtained by Krawtchouk moment invariants are clustered by kernel fuzzy C-means (KFCM) to realize fabric defect detection.Compared with PCNN, Gabor wavelet and the method combined Gabor and fuzzy C-means (FCM),this method has been greatly improved and is an effective method of automatic fabric defect detection.Finally, a fabric defect detection method based on local binary patterns (LBP), Krawtchoukmoment invariants and wavelet support vector machine (SVM) is implemented. The texture feature offabric defect image extracted by LBP and shape feature extracted by Krawtchouk moment invariantsare synthesized and as the input of wavelet SVM for defect detection. The experimental results showthat this method can detect defect regions effectively and the edge is clear.