Font Size: a A A

Image Vectorization Based On Hierarchical Optimization

Posted on:2020-12-20Degree:MasterType:Thesis
Country:ChinaCandidate:Y P WangFull Text:PDF
GTID:2428330590496834Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
Image vectorization is a common representation method that converts raster image into vector image.Image vectorization is also an important research content in digital image processing.How to select appropriate vectorized graphics to edit images so that the original image can be described as accurately as possible from the rendered images of vectorized graphics has always been the core issue of image vectorization,and there are some research difficulties in different image vectorization algorithms.So when using vector graphics to represent images,its basic task is how to efficiently solve the above problems.With the improvement of current technology,the research of image vectorization technology has practical significance.Aiming at the research and analysis of related image vectorization representation algorithms,a new image vectorization representation method based on hierarchical optimization is proposed in this paper.Firstly,the given image is filtered and vectorized with high quality image.Secondly,thin-plate spline fitting is used to transform discrete pixel functions into continuous functions.Then the initial number of mesh blocks is given according to the complexity of the image.Thirdly,the gradient mesh represented by B-spline surface is improved.In this paper,the gradient mesh represented by B-spline is used to vectorize the R,G and B channels of color image,and the gradient mesh satisfying the optimization error is found layer by layer.When optimizing the energy function,the approximation error of the image is reduced by refinement of the mesh.Finally,Levenberg-Marquard algorithm is used to solve gradient mesh alternately and iteratively to find the optimal gradient mesh.For each step of solving gradient mesh,the method of optimizing one point to fix other points is used,so the mesh points obtained are local optimum.The test results show that the gradient mesh represented by B-spline parametric surface has better shape control in rendering image,and the fitting error of gradient mesh represented by B-spline surface is smaller under the premise of the same number of mesh vertices and surface patches.
Keywords/Search Tags:Image Vectorization, Image Complexity, Thin Plate Spline Fitting, Levenberg-Marquard Algorithm, Surface Representation
PDF Full Text Request
Related items