Intelligent Recognition Of Embroidery Stitches Based On Multi-feature Fusion

Traditional embroidery is an intangible cultural heritage of the Chinese nation,which is energetically promoted by the state.At present,most research focuses on the cultural deposits of embroidery and its application in fashion products.There is relatively little research on embroidery stitch techniques,especially in identifying embroidery stitch techniques from the perspective of images.The stitch techniques recognition of embroidery works mostly relies on experience to distinguish,and there are relatively few intelligent recognition methods,which is adverse to the digital inheritance and protection of embroidery art works.This article summarizes the commonly used embroidery stitch techniques based on the theory of embroidery and stitch techniques,and analyzes their features from the perspective of stitches.It improves the traditional Snake algorithm and proposes the HC-Snake algorithm for object segmentation of embroidery stitch technique samples.The shape,texture,and color features of the target samples are extracted,and a feature database Feature is established to store embroidery stitch technique features.An intelligent recognition model for embroidery stitch techniques is created.The innovation points of this article mainly include the following parts:On the basis of analyzing the characteristics of embroidery stitch techniques,a more systematic classification of embroidery stitch techniques was carried out,and a stitch technique model was constructed.On the basis of analyzing the characteristics of embroidery stitch techniques,three major stitch technique classification methods were proposed:point stitch technique,line stitch technique,and block stitch technique.Feature analysis was conducted on various stitch techniques of the three categories,and a needle technique model was constructed.And the shape features of the three types of stitch technique images were used to identify the stitch technique categories of stitch,while the texture and color features of stitch technique images are used for the recognition of specific stitch technique categories.Improved the traditional Snake algorithm that relies on manual point selection,and proposed a HC-Snake contour recognition algorithm that automatically selects control points for object shape feature extraction.On the basis of enhancing the texture details of embroidery stitch technique images,the traditional Snake algorithm is improved:feature points are automatically selected based on Harris corner detection,combined with Canny operator edge detection,corner points are filtered to form HC-Snake algorithm contour control points.Using the minimum bounding box algorithm to extract shape features,using the gray level co-occurrence matrix to extract texture features of the target image,and using color moments to extract color features,a total of 25 feature parameters were extracted.By analyzing the variance of texture and color feature data,nine feature parameters,CON_sd,IDM_sd,E_a,E_sd,ENT_a,ENT_sd,COR_sd,SM,and TM,were determined as classification feature indicators to establish a feature dataset.A highly targeted feature access database feature has been designed based on the different characteristics of embroidery stitch techniques.Based on the relational database My SQL,combined with the embroidery stitch characteristics of this study and the subsequent requirements for the convenience of reading and writing the stitch eigenvalue,the logical structure table of the embroidery stitch’s shape characteristics,texture characteristics and color characteristics was designed to complete the construction of the database.A embroidery needle intelligent recognition model combining SVM and BP neural network is proposed.Based on the SVM algorithm,a stitch recognition algorithm is designed using the 16dimensional shape features of the embroidery stitch method in this study as parameters.Then,based on the BP neural network,a specific stitch recognition algorithm is designed using the 9 dimensional texture and color features of the embroidery stitch method as parameters.The two algorithms are fused to form a model system for accurate recognition of large and specific categories of embroidery stitch.By constructing the confusion matrix of recognition results,calculate the accuracy and recall rate of the intelligent recognition model,and confirm the reliability of the model.Then,an example verification was conducted on the recognition model.Finally,the concept of the application software for the model is proposed,and the interface design of the application software is carried out.In summary,the HC Snake algorithm proposed in the paper can effectively improve the shortcomings of traditional Snake algorithms in manually selecting control points,making it more intelligent;the embroidery stitch classification model established by the research institute can be used for needle recognition of ordinary network embroidery images,improving the accuracy and applicability of the classification model.This research achievement helps to promote the digitization of traditional embroidery art works,thereby achieving the digital inheritance and protection of embroidery arts.