Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (17): 208-213.DOI: 10.3778/j.issn.1002-8331.1704-0334

Previous Articles     Next Articles

Portrait paper-cut synthetic method based on facial features

WANG Xiaoyu, ZHAO Yan, LIU Zeyun, FENG Jun   

  1. School of Information Science and Technology, Northwest University, Xi’an 710127, China
  • Online:2018-09-01 Published:2018-08-30

基于五官特征的人脸剪纸生成方法研究

王晓宇,赵  妍,刘泽云,冯  筠   

  1. 西北大学 信息科学与技术学院,西安 710127

Abstract: Portrait paper-cut is supposed to reproduce the vivid image details. Therefore, based on facial features and image morphing algorithm, a two-phase synthetic method of paper-cut is proposed. Firstly, it collects the face paper-cuts by professional artists, decomposes them into facial components, extracts their geometric features, and establishes a digital paper-cut database of facial features. It demonstrates a paper cutting synthesis toward the target facial images. The first stage is to demarcate the image feature points of the target face, divide the facial parts, and extract geometric features from each component. After that, it calculates the similar measurement between the target facial features and the corresponding components in the paper-cut database based on the geometric features and the shape context, a higher similarity of paper-cut parts can be selected by fusing the geometric features and the shape context features, so that after splicing, an initial personal portrait with a paper-cut style can be obtained. The second stage is to deform the face paper-cut generated in the first stage with the Thin Plate Spline(TPS) deformation algorithm to obtain the final personal portrait with a paper-cut style. It has been shown by multiple visual experiments that this method can be expected in terms of getting a satisfactory personal paper-cut portrait.

Key words: paper-cut, geometric feature, shape context, image deformation

摘要: 人脸肖像剪纸应该重现生动的图像细节,为了实现这一目标,提出了一种基于五官特征与图像变形算法的两阶段人脸剪纸合成方法。收集艺术家的人脸剪纸创作,分割五官部位并提取各组件的几何特征,建立数字化五官剪纸数据库。对目标人脸图像进行剪纸合成:在第一阶段,标定目标人脸图像的特征点,分割其五官部位,并提取各部位的几何特征,之后分别计算目标人脸五官与剪纸数据库中各对应组件基于几何特征和形状上下文特征的相似性度量值;通过融合几何特征和形状上下文特征,选择匹配相似度较高的剪纸部位,拼接得到初步的人脸剪纸图。在第二阶段,采用薄板样条(Thin Plate Spline,TPS)变形算法对第一阶段合成的人脸剪纸图进行变形,得到最终的剪纸图像。通过多人视觉测评实验,结果表明运用该方法得到的人脸剪纸图能够达到较为满意的效果。

关键词: 剪纸, 几何特征, 形状上下文, 图像变形