Research On Recognition And Matching Of Collar Types In Garment Styles

The design of traditional clothing structure mainly depends on the completion of professional platemakers.Its own limitations cause it to be time-consuming,inefficient,and costly,and the quality of platemaking depends entirely on the professional level and experience of the platemaking personnel,resulting in a large subjective influence.In addition,the change of personalization concept is prompting the apparel industry to develop in the direction of small batches and multiple varieties.Improving the "quick response capability" of apparel companies has become the key to improving their competitiveness.Therefore,it is urgent to find a method to achieve rapid plate making.This subject takes clothing style drawings as research objects.Image processing and recognition technology was used to retrieve parts from the sample library that were similar in style to the target clothing part(taking the collar as an example).Users only need to finetune the corresponding structure diagrams to get new samples,which can effectively reduce unnecessary repeated operations,improve efficiency,and realize resource sharing.At the same time,this paper also provides a new idea for automatic style recognition,software development and application.The main research contents of this topic are as follows:Firstly,the collar style sample library is constructed,which contains 8 types of collars commonly used in clothing,which are round neck,V-neck,square collar,stand-up collar,shirt collar,flat collar,connection collar and closure collar.And each type includes 60 images.The advantages and disadvantages of common image preprocessing methods are compared and analyzed,such as image graying,sharpening,edge detection,morphological processing,and image segmentation.Based on this,a suitable image preprocessing scheme is formulated and the algorithm is implemented on the MATLAB platform.Then,a shape feature extraction algorithm based on complex network is proposed based on the image features of the collar style.First,an initial network is constructed with contour points as nodes and a multi-range threshold evolution is performed.Second,the shape feature description operator is constructed by extracting the maximum degree,average degree,average joint degree,entropy,energy,average clustering coefficient,and average path length of each sub-network after evolution.Then,the support vector machine and BP neural network were used to identify and classify the collar styles.Among them,the average overall classification accuracy based on BP neural network is 95%;the average overall classification accuracy based on SVM is 98%.In summary,the classification effect is quite satisfactory.In addition,its anti-noise performance is verified by adding different levels and types of noise.The results show that the method in this paper is suitable for image recognition with a certain noise density,and it still has a good classification effect when a certain level of noise is added.Meanwhile,the algorithm in this paper is compared and analyzed with Hu's moment invariant and HOG feature extraction algorithms.The experimental results show that the classification accuracy of the algorithm in this paper is the highest,and its classification result is relatively stable,whether it is in the overall classification of the sample or the average classification of each category.In summary,the complex network-based feature extraction method proposed in this paper is suitable for the identification and classification of collar styles.Finally,based on the collar component feature library constructed in this paper,image matching was performed using Euclidean distance and Manhattan distance as similarity measures,and recall and precision as effective measures.The results show that higher recall and precision can be obtained by using Euclidean distance.Among the matching results of 8 images(8 categories)based on Euclidean distance measure,the correlation coefficient of 4 images is 1;the correlation coefficient of one image is 0.99;and the correlation coefficient of one image is 0.89.In summary,the result of style matching is ideal,which can meet the purpose of sharing the collar structure drawing.