Research On Garment Drawing Pattern Recognition And Its Application In Retrieval

With the development of garment CAD technology,more and more style drawings and corresponding pattern files have been saved.The rapid reaction of garment drawing and structure pattern,the full use of repeated styles and pattern components are the demand of enterprises under the current big data and intelligent background,save the cost and speed of the garment design and plate making process,and thus improve the production efficiency.Therefore,how to achieve the scientific and effective convergence of garment drawing and structure pattern and how to make full use of duplicated style parts and apply them to clothing intelligent design become an important research direction for the development of digital apparel design technology.Based on image processing and pattern recognition technology,this paper realizes the preprocessing,feature extraction and description,component classification and recognition of garment drawings.At the same time,the design of the pattern recognition implementation scheme is applied to the whole theoretical framework of the pattern retrieval and intelligent design system.The main contents are as follows:Firstly,8 different types of collar parts are made,in which the number of 60 parts and the overall pattern of 100 pieces are used to provide training and test samples for the experiment.Secondly,the preprocessing,feature extraction and classification recognition of the drawing are carried out.In the preprocessing,the pattern is sharpened and binaryzation,the result of Laplace sharpening is better,and the binaryzation is ideal when the threshold is 0.6.The obtained image is segmented by the region growing method.In comparison,the edge detection based on the Canny operator is more effective.Finally,the contour point set of the parts that detect the edge is sequentially parsed and the component vectorization based on the Hough and Bezier curves.After the program runs,the effect of segmentation and vectorization is good.In the extraction and description of the features of the drawing,the chord feature matrix(CFM)method is used.When the contour points areselected 32,64 and 128,the chord characteristics of the scale selection for 4,16,32,64 are well described,and the convex and concave points of the collar are well described.In the component classification and recognition,normalized the extracted three chord eigenmatrix and the Fourier transform is selected to normalize the style rotation characteristics.When the number of Fourier low frequency coefficients is different,the parts are reconstructed.It is found that when the number of low frequency coefficients is18,the original image is basically restored,and support vector machine(SVM)classification is used.Nearest neighbor(1NN)classification identifies the components to be identified.The number of contour points is 64,5 different scales,and 16 Fourier coefficients of low frequency are used for classification recognition experiments.The results show that the combination of CFM feature extraction and SVM and 1NN recognition classification method is very effective,and the integrated recognition accuracy is 95.6%,and the target can be identified according to the degree of the difference.At the same time,the control variable method is used to analyze the results of the combined chord feature descriptor(CFM)and SVM and 1NN.When the feature extraction of the original image of the salt and pepper noise is extracted,the CFM feature extraction selected in this paper is robust;and under the same classification and recognition method,the recognition rate based on Hu invariant moment and curvature feature is lower than CFM,which are 82.5% and 88% respectively,but the spending time of the entire Hu invariant moment is short;Based on BP neural network classification and recognition rate is 62%,which is lower than the SVM and nearest neighbor classification methods in this paper.Finally,compared with the SURF identification method,this paper obtains the advantages of the whole style picture identification.Finally,the importance of applying the garment drawing recognition to pattern retrieval and intelligent design system is studied,and the theoretical framework diagram of the CAD system of the whole smart garment design is proposed.