Towards Error-controllable Image Vectorization Based On Subdivision Surfaces

Image vectorization converts raster images into vector representations.Due to the diversification of display devices and improvement of their resolution,the advantages of vector images are becoming more and more prominent.This paper proposes an error-controllable framework based on the subdivision surfaces image vectorization algorithm by Liao et al.Our system works as follows.It first employs an edge drawing method to detect feature lines of the image.In order to reflect the color discontinuity on both sides of the feature lines,we propose two strategies to obtain two feature lines representing the color values on both sides.The one makes use of a contour tracing algorithm to get the other feature line on one side of the original feature line to form two separated feature lines,the other directly split the original feature line into two overlapped feature lines.We then construct a dense initial mesh,which respects to the feature lines,and mark these feature lines in the mesh.Following that,we simplify the initial mesh with an improved the QEM simplification algorithm in order to get a good quality base mesh.In the final step,we employ Loop subdivision surfaces with sharp feature settings to fit the color height field of the image.During subdivision surface fitting,we repeatedly insert new control points into the base mesh in order to reduce the fitting error between the reconstructed image and the original image.From experimental results we find the proposed framework can obviously improve the quality of the reconstructed images compared to the original fitting results,and achieves error-controllable fitting to some extent.