基于NMF和LS-SVM的肖像漫画生成算法研究

Facial Caricature Generation Based on NMF and LS-SVM

提出一种基于非负矩阵分解(NMF)和最小二乘支持向量机(LS-SVM)的肖像漫画生成算法。在训练阶段,利用非负矩阵分解来对夸张特征空间数据降维,运用最小二乘支持向量机(LS-SVM)统计学习夸张漫画与人脸之间的关系,建立形状夸张模型。在应用阶段,利用AAM算法提取人脸特征点,形状夸张模型计算出相应的漫画特征点数据,经过图像变形和风格化即可得到最终的肖像漫画。算法实验表明,该算法可以合理地夸张主要特征并避免过度变形。

A LS-SVM and NMF based caricature generation method is proposed.In the training phases,the set of exaggerated caricature and original image pairs,manually labeled feature points both on the original images and the exaggerated caricatures are started with,and Nonnegative Matrix Factorization（NMF） for data dimension reduction are used.To build the shape exagge-ration model,LS-SVM to learn the relationship is applied between caricature and facial features.At runtime,for a given image,the face shape is extracted using AAM.Then,the shape exaggeration model is employed to generate the exaggerated shape.This algorithm can exaggerate the main features and avoid excessive deformation.