Development And Implementation Of A Multi-task-driven Vectorization System Based On Traditional Ethnic Motifs

Nowadays,there is a lack of research on vectorization of traditional ethnic motifs,the existing algorithms are difficult to extract the motifs from the carriers,the algorithms are not robust,and the output results cannot be finely controlled,so an efficient vectorization platform needs to be developed.This paper takes traditional ornament patterns as the research object,carries out the research of vectorization solution,innovatively proposes a multi-task-driven vectorization algorithm,and finally builds a vectorization system for traditional ethnic ornament patterns,The research content and innovation points mainly include:(1)Based on the single-task-driven idea,a special algorithm is used to optimize each type of material.There are a large number of binary images in the material library,so this paper adds a binarization pre-processing process based on the existing binary region algorithm,and also optimizes the original algorithm to propose a new polygon contour tracking algorithm,which can reduce the time complexity of path to polygon representation from the original O(n2)to O(n)under the optimal case.(2)The extraction scheme of the ornament pattern is proposed.Since there are many carriers of ornament patterns such as clothing and artifacts,the ornamentation on the fabric is disturbed by the fabric texture,and it is not very realistic to vectorize these patterns directly.For this reason,a line extraction algorithm based on edge tangent flow field(ETF)is used,which performs well in the ornamentation material and can extract the complete decorative lines of ornament patterns.(3)A vectorization algorithm based on hierarchical clustering is proposed based on a multi-task-driven vectorization solution.The algorithm solves two problems:the binary algorithm cannot handle color images;the degree of detail of vectorization results cannot be precisely controlled.To address the first point,we propose to use hierarchical clustering to extract regions of color images,and each layer of regions is regarded as a binary image element.The algorithm is highly optimized,and even though the number of regions obtained by hierarchical clustering is large,the performance of the algorithm is still excellent.To address the second point,three metrics of accuracy,continuity and simplicity are proposed.Using the characteristics of hierarchical clustering tree,users can adjust the parameters to obtain the ornament outline with less number of regions and better simplicity,and also get the vector map with better accuracy and more details.(4)Based on Java backend framework and Web front-end technology to realize traditional ethnic ornament pattern vectorization platform.The system design is reasonable and allows third-party vectorization algorithms to be accessed after authentication,which provides convenience for the integration of subsequent excellent vectorization algorithms.The system interface is simple and easy to interact with,providing one-click vectorization function,and the algorithm is highly optimized for good platform performance.