Research On Automatic Recognition Of Fabric Weave Patterns Based On Digital Image Processing

Experts or professional workers often analyze fabric structure via their experience, magnifying glass and other auxiliary equipments in textile industry. Although high accuracy is achieved by this way, traditional detection is tedious and time-consuming. Meanwhile, high centralized and tension state of inspectors during long processing has influence on their physical health. Therefore, image processing and pattern recognition is adopted to classify fabric texture and compute warp and weft density for realizing the automatic and intelligent production of textile industry. In this paper, plain, twill and satin woven fabric are put as research objects. Structure analysis of woven fabrics is based on classification and identification, respectively.On one hand, automatic classification of fabric structure is realized. Firstly, fabric image is preprocessed by median filter and bimodal Gaussian function specification algorithm in order to filter noise and improve contrast. Then, the two approaches which are local binary pattern and gray level co-occurrence matrix are applied to extract the local and global texture features of fabric image. At last, an appropriate BP neural network classifier based on Levenberg-Marquardt(L-M) algorithm is used to training and testing the feature vector in order to achieve the automatic classification of three basic woven fabrics(plain, twill and satin weave). Finally, compared with GLCM method and LBP method, the fusion of the two feature vectors obtains the best classification result(99.33 %).On the other hand, woven fabric density is inspected automatically. Firstly, De-nosing and contrast improving of fabric image is achieved by histogram equalization and filtering. Secondly, fabric images that local contrast has been enhanced are multi-decomposed and reconstruction by biorthogonal wavelet. The warp and weft information of fabrics is separated. To obtain warp and weft images with clear and arranged yarns, sub-images are processed via morphology and other methods. Thirdly, density of the warp and weft yarns could automatically been calculated through computer programming for using the relationship between resolution and pixels.Automatic classification and recognition of woven fabric structure are achieved effectively. It is not only reduces manual errors detected by the presence, but improves the efficiency and accuracy of detection. Meanwhile, this research has great theoretical significance for textile products reproduction and better application.