Fabric Defect Detection And Recognition Algorithm Based On Image Segmentation

Fabric defect detection and recognition has been studied based on imagesegmentation techniques in this thesis. Fabric defects is an important factor affecting thequality of the fabric, so the fabric defect detection is an essential part of the productionprocess. In order to improve the detection efficiency and accuracy, we should developeffective detection algorithm applied to the automatic detection system. Digital imagetechnology can reflect the two-dimensional information body quickly and intuitively. Thetechnology have a good application in the military, geological surveying, satelliteimagery and other fields because of those advantages and in accordance with thetechnical requirements of the fabric detection.This thesis describes the two types of image de-noising technology: low-pass filteringnoise and wavelet transform denoising in order to improve the image quality of image forthe foundation of segmentation. Secondly our thesis discusses two common methods forimage segmentation namely the threshold segmentation method and the fuzzy C-meansegmentation method. The idea of the former is dividing the image into two categoriesnamely the target and the background using a reasonable threshold. This method issimple and fast. This method can get good result on fabric image with relatively simpletexture. Fuzzy C-mean clustering is a common method for image segmentation. Thismethod get segmented image with the all pixel belong to C categories after iterativecomputation. The fuzzy C-mean clustering overcome the shortcoming of "hard"classification and get good results by introducing membership function, but the processof segmentation is sensitive to noise and the speed reduces seriously when the datasamples increases.Then, we propose a fabric defect detection algorithm based on improved thresholding, the algorithm is based on the idea of linear discriminate analysis. We improvevalley-emphasis method. The objective of the improved method is the optimal thresholdfor image segmentation not only considering the between-class variance but also theinter-class variance. When the between-class variance is as large as possible and theinter-class variance as small as possible, we obtain the optimal threshold forsegmentation. We also propose a fabric defect detection algorithm based on optimizedFuzzy C-Means method. Firstly we preprocess the fabric image with wavelet transform,then convert the sample data from pixel space to histogram space to improve the speed ofalgorithm. Experimental results show that the two improved methods obtain better resultswhen dividing fabric defect image.Finally we design the BP neural network to recognize the fabric defect and test thenetwork using the test samples and the effect is good.