The Techniques and Evaluation Method for Beautification of Handwriting Chinese Characters Based on Cubic Bézier Curve and Convolutional Neural Network

Abstract

This paper presents a method to beautify Chinese characters and a way to evaluate the beautification result. In order to make handwritten Chinese characters more in line with the aesthetic standards of Chinese characters, 52 Chinese characters were selected as experimental data. These data covered 33 standard strokes and 19 typical structures of Chinese characters. The handwritten Chinese characters were beautified mainly from two aspects-the global adjustment and the elimination of jitter. Firstly, the two-dimensional (2D) data points set is extended into three-dimensional (3D) space. Then the Gaussian Mixture Model (GMM) is established for the data set, and the layout of handwritten Chinese characters is adjusted by point set registration algorithm. Secondly, according to the properties of the cubic Bézier curve function, detect the jitter of each strokes, and eliminate the jitter by interpolation algorithm. The evaluation of the results after beautification has always been limited to subjective evaluation. This paper attempts to combine the evaluation of beautification result with machine learning methods. Handwritten Chinese character recognition (HCCR) is used as the tool. Experiments show that the overall layout and jitter of handwritten Chinese characters have been adjusted and deleted, and the evaluation of handwritten Chinese characters beautification results has its research significance.