Fabric Defect Detection Method Based On Texture Enhancement

The appearance defect of fabric is an important factor which influences the quality ofproducts, and defect detection becomes increasingly a significant task to control the qualityin the industry of textile. It is of high precision and efficiency to detect the defects of textileusing the image processing technology compared with the traditional artificial detectionmethods. The texture detection methods for fabric are studied deeply, and the main work inthis thesis is as follows:(1) The characterization methods for texture images in the domain of frequency andtime are summarized and the feature extraction methods for fabric using gray levelco-occurrence matrix are researched. For the texture characteristics possessed specificallyby fabric, the deviation between textures caused by noise in the areas of defects andbackground is analyzed.(2) A fabric enhancement method based on PCA-NLM is proposed, for the NLMmethod, some redundant information caused by noise exists in the image information whichis used to compute the neighborhood window similarity, the denoising and parametersestimation are carried out simultaneously when using PCA method, as a result, the texturestructural information of fabric images is preserved efficiently and which also benefits tothe following defects detection.(3) A fabric defect detection method based on texture enhancement is proposed, theclass separability between defective texture and non-defective texture is analyzed, thedetection discrimination of gray level co-occurrence matrix features in the defectiveareas is improved and the texture detection accuracy is increased.