应用三角形坐标系的风格化肖像生成算法
A Generating Method of Stylized Portraits Using Triangle Coordinate System

针对现有肖像漫画化处理方法中存在的依赖训练样本以及算法复杂度高的问题,为了在准实时系统中实现应用,提出了一种应用三角形坐标系的风格化肖像生成算法。该算法自动提取输入的实验人脸特征,将其与参考人脸比较得到个性特征,并通过夸张个性面部特征得到风格化肖像特征;分别基于实验人脸特征和风格化肖像特征建立三角形坐标系,在2个三角形坐标系之间实现图像变形生成风格化肖像。该算法无需训练样本,计算简单、全自动化,变形灵活且可预期,可快速生成风格化肖像。实验结果表明,该算法能自动提取并夸张实验人脸的个性面部特征,且算法执行效率高,可在0.5 s以内生成效果较为生动的风格化肖像,表明该算法的综合性能良好,可在准实时的虚拟现实、增强现实和肖像绘制机器人等系统中应用。
A generating method of stylized portraits based on the triangle coordinate system is proposed to solve the problem of the dependence on training samples and the high complexity of the existing algorithms, and to realize application in quasi real？？time systems. The proposed algorithm automatically extracts the features of an input face and calculates the personalized facial characteristics by comparing the features with those of the reference face. Then features of its stylized portrait are determined by exaggerating the personalized facial characteristics. Two triangle coordinate systems are established based on the features of the input face and the stylized portrait respectively, and the stylized portrait is generated by image morphing between these two systems. The algorithm does not need training samples, and is simple, automatic, flexible and predictable, and it can quickly generate stylized portraits. Experimental results show that the algorithm automatically extracts and exaggerates personalized facial characteristics of an input face, and produces a vivid stylized portrait in less than 0？？5 second. The algorithm has good comprehensive performance, and can be applied to the systems of quasi real？？time virtual reality, augmented reality, and portrait rendering robots