An Investigation Of Yarn-dyed Woven Fabric Pattern Recognition Based On Dial-side Imaging Technology

The weave pattern is an important parameter of woven fabrics and plays a decisive role in fabric appearance analysis and physical properties evaluation,thus fabric pattern recognition is an indispensable part in textile quality test.The traditional method of weave pattern recognition manually is not only time consuming and low efficiency,but also subjectively effected by the knowledge and experience of inspectors.In recent years,more and more researchers attempte to utilize computer image processing technology to do textiles quality test.For fabric pattern texture identification,it is mostly based on single-side image analysis,and difficult to accurately identify the fabric structure and color information.Therefore,this paper proposed a method of recognizing yarn-dyed woven fabric pattern based on dual-side imaging technology.Firstly,the establishment of the dual-side image acquisiton system and the workflow of the dual-side imaging is introduced.The reference makers on sample holder and Radon transform are used to match the face-side image with back-side image on the pixel level.Then,the color images of the woven fabrics are transferred from the RGB to the CIE-Lab color space,and the intensity information of the image is extracted from the L component.In order to reduce the difference of yarn brightness in grayscale image and enhence the yarn structure information,the Butterworth high-pass filter is utilized to remove the low frequency information in image frequency domain.Subsequently,three types of image fusion algorithm are developed and utilized to merge the dual-side images: the weighted average method,wavelet transform method and Laplacian pyramid blending method.The fusion effcacy of each method is evaluated by three evaluation indicators,the results show that the fusion method based on Laplacian pyramid blending is better than others.Then,the fast Fourier transform is uesd to convert fusion image from spatial domain to frequency domain,and the frequency peak points contain information of yarn periodical arrangement are extracted.The inverse fast Fourie transform is used to reconstruct of the warp and weft yarns images,and then yarn density can be calculated according to them.Besides,the noise in fabric image has been removed by using adaptive filter,and the image contrast has been enhence with the aid of histogram equalization.Subsequently,the gray projection is utilized to accomplish yarn location and segmentation for cross-point grid initialization.The grid adjustment method based on local warp intensity has proposed to adjust the grid position by locating the local minium of warp intensity,to make the grid contains complete edge information of cross-point.After that,a cross-point classification method based on template matching has been proposed: firstly,the template images of warp cross point and weft cross point are constructed according to gray features of two kinds of cross-point;secondly,two feature parameters of cross-point image,horizontal gray-level variation value and vertical gray-level variation value,are extrated to match with values extracted from template image for cross-point classification;finally,weave diagram is obtained according to classification results.Besides,the dual-side weave diagram is used for error classification detection of cross-points,and the k-nearest neighbor algorithm is utilized to resive the weave diagram.After the color feature parameters of each cross-point have been extracted,dyed warps and dyed wefts are merged respectively through color feature parameters and dual-side weave diagram.Finally,a self-adaptive K-means algorithm has been used to carry out the color clustering of dyed yarns for color pattern diagram construction of yarn-dyed woven fabrics.