Application Of Local Binary Pattern Algorithm In Fabric Defect Detection And Classification

Defects on textile surface have great impacts on the value of textiles. To detect and repair defects occurring on the production line is essential. At present in domestic factories the mainstream approach to detect defects is artificial inspection.Because the inevitable shortcomings of artificial inspection, such as slow detection speed,detection accuracy can not be guaranteed, easily influenced by subjectivity, as well as the great damage to inspectors’ eyes and body, it will inevitably need to study and propose the automatic testing system complies with cloth industrial production and development.It is bound to study and propose an automatic fabric inspection system to meet the industrial production and the development.Local Binary Pattern(LBP) is proposed as a texture description operator. Because of its highlighted ability of describing texture local information, strong ability of classification, high efficiency of calculation, and it not subject to monotonic gray-scale changes, LBP is widely used in texture classification, facial feature recognition and image retrieval analysis etc al. by the vast number of domestic and foreign scholars in recent years and the effect is relatively outstanding. Thus begin the design of automatic detection system with the help of the good application of LBP operator in various fields.Algorithms of detection and classification of textile defects is the soul part of the automatic detection system. In the thesis, some researches are done based on LBP, such as:1. The introduction of LBP operator. In this part detail introduce the production and the statistical law of LBP operator. Moreover, summary the improved methods in recent years after deep study of LBP algorithm and give the final method to describe texture feature values.2. For the textile defect detection, using improved LBP algorithm to design the system software. Firstly use LBP operator to extract the feature of the whole defect free fabric images, then divide the defect free fabric images into small detection windows and obtain the features of each window. Calculated to the similarity of the two parts of features and select the max as the threshold. And then process the defect images as thesame steps. Compare the similarity and the threshold to get the defects area, so as to detect the defects.3. Improve the LBP algorithm by add two new parameters which is the meanmg and variancesg of the pixel gray values in the neighborhood. With the new parameters as threshold to code the center pixel, it can reduce the loss of useful information compared with the basic LBP. Combined with the good clustering ability of SOM neural network, the features are input into the SOM neural network to be classified into two parts. Then defect information of the image can be segmented, the purpose of defect detection can be achieved at the same time.4. To achieve the goal of the classification of defects, an algorithm combined with LBP and gray level co-occurrence matrix(GLCM) is put forward. First of all, the local feature information of images is extracted by LBP algorithm. And then through the GLCM to extract the overall texture information of the image. Finally the two parts of feature information are structured as a whole to be put into the improved BP neural network input. And the trained network can be used to classify and identify the different types of defects.