Automatic Extraction Of Warp Knitted Fabric Pattern

With the rapid development of warp knitting, our country is in the transformation form a big warp knitting country to a powerful warp knitting country. But we still rely on artificial to complete production and testing work which is time-consuming and strong subjectivity. In general to sample design and test pattern whether its shape, we all need to segment the pattern.But in the design of fabric pattern, many enterprises still stays in the traditional jacquard graph method which already cannot meet the requirements of the development of modernization, severely reduces our country’s international competitiveness. In this paper,computer vision and image processing technology is applied to textile areas. Expected to implement the intelligent of the division of fabric patterns, so as to improve the design efficiency of fabric patterns.This paper introduced the research status firstly. The main object is jacquard warp knitting fabric patterns, generally due to the warp knitting jacquard fabric surface will form a thin, thick and mesh effect of three kinds of fabric pattern. And through the three effect and the different combination of pressure and lining weft yarn, thus forming the patterns of different texture. This paper is mainly based on color, texture and spatial characteristics for the fabric pattern division.Introduced the texture segmentation algorithm, for the warp knitting jacquard fabric patterns, this paper proposes a new texture feature extraction method. The dual tree complex wavelet transform is combined with the energy matrix to describe the local texture of fabric image. After dual tree complex wavelet transform, image of fabric pattern has six direction subband at each scale. Its more direction provides guarantee for image local details of texture feature extraction.This paper mainly chosen the color, texture and spatial characteristics, and used the classical PCA algorithm for optimization of feature dimension. Then we proposed Gaussian mixture model(GMM) segmentation algorithm which based on EM algorithm. The EM algorithm is a kind of segmentation method based on statistical pattern recognition which has the characteristics of the rapid and broad adaptability. EM algorithm converts pattern extraction parameters of maximum likelihood estimation problem, according to the each pixel belong to different size of Gaussian posterior probability sorting. Pixels are classified as itsvector parameter value of the maximum a posteriori probability in class, then each pixel of image annotation, which complete the fabric image segmentation. Experiments showed that the method of image segmentation effect is good which can very good to segment the fabric patterns at all levels. The method has a high accuracy, high speed and has a certain practical value.