Study On The Influence Of Woven Fabric Structure On Texture Representation In Dictionary Learning

The warp and weft yarns intertwined on the surface of the fabric to form a clear texture according to a certain rule.The geometric shape of the yarn in the fabric called the structure,and the different structures of woven have different textures.The texture contains information such as warp and weft density,structure,fabric flaws,etc.The characterization of woven texture is one of the hotspots in recent years.In this paper,the dictionary learning algorithm used to represent the woven fabric texture.The dictionary obtained by the algorithm can contain the main texture information of the woven fabric,allowing the dictionary to representation the original image.The K-SVD dictionary learning algorithm used to reconstruct the original image of the woven fabric,and the similarity between the reconstructed image and the original image compared to analyze the influence of fabric structure on the dictionary learning algorithm and texture characterization.The main research contents of this paper are as follows:(1)Weaving plain,twill 1,satin,reinforced twill,compound twill,entwining twill,reinforced satin,barley corn weave,square check weave,crepe weave1,mesh weave,honeycomb weave,huckaback weave,horizontal cord weave,vertical cord weave woven fabric used as samples,and the warp and weft density are about 350/10 cm,and the texture image collected and pretreated.(2)The K-SVD dictionary learning algorithm used to construct the optimal dictionary and the sparse coefficient to reconstruct original image,and the original image reconstructed using the dictionary and sparse coefficients to get the reconstructed image.The size of the experimental sample is selected,combined with the number of tissue cycles of the sample,the density of the sample,and the SSIM value and the MSE value of the reconstructed image quality evaluation index,and the selected experimental sample size is 128×128 pixels.(3)Using K-SVD algorithm to characterize the woven fabric texture,the algorithm parameter sub-window size,sparsity and dictionary atom number studied to affect the similarity between the reconstructed image and the original image.The first is the preference of the parameters: the experiment carried out by changing the values of the sub-window size,sparsity,and the number of atomic numbers of the dictionary.The results show that the larger the sub-window size,the larger the reconstruction error,and the smaller the sub-window,the easier it is.The randomness of the texture difference reflected;the larger the sparsity,the more the number of dictionary atoms used to express the subsample,the smaller the reconstruction difference,but it takes a long time,and for the woven texture,there are fewer dictionary atoms.The image can reconstruct;the number of dictionary atoms is large,and the more the amount of texture information included,the reconstructed image has good approximation,but the dictionary may contain repeated texture information.Considering the sample tissue circulation and the reconstructed image quality,the sub-window size is 8×8 pixels,the sparsity is 4,and the dictionary atom number is 500.(4)After determining the experimental parameters,the effect of woven fabric structure on dictionary learning texture representation studied.According to the results of evaluation index SSIM value and MSE value,the texture information of woven fabric image is different due to the difference of structure,and the error between the reconstructed image and the original image is inconsistent.The experimental results show that the fabric has small tissue circulation,long and more float lines,the more obvious texture,good regularity,simple image texture information,and the reconstructed image approximates the original image.The tissue circulation is large,the long and short float lines interlaced,the surface of the fabric formed with concave and convex effects,the image texture information is complex,and the reconstructed image quality is poor.(5)The more complex the woven fabric texture information,the more dictionary atoms needed to represent it.Thus,according to the complexity of texture information,the number of dictionary atoms used as the standard,and the samples divided into three categories.The first category is barley corn weave,entwining twill,crepe weave 1,mesh weave,huckaback weave,honeycomb weave,and the number of such dictionary atoms is large.The second category is to reinforced twill,compound twill,reinforced satin 2,reinforced satin 4,vertical cord weave,such dictionary atomic number is in the middle.The third category is plain,twill 1,satin,square check weave,horizontal cord weave.This type of texture can representet by a small number of dictionary atoms.The above classification verified by using twill 2,diamond twill and sputum tissue 2.The experimental results show that the classification results based on the dictionary atom number are related to the classification of structure characteristics,and the classification results are accurate.In summary,this paper reconstructs woven texture image based on dictionary learning algorithm,and focuses on the influence of woven fabric structure on texture representation,which provides some reference value for dictionary learning to characterize woven texture and classification of woven fabrics.