应用于草图的人脸素描合成方法  

Face Sketch Synthesis Based on Rough Drawing

作　　者：梁浩然[1] 宋明黎[2] 梁荣华[1] 蒋莉[1] 王子仁[1] 毛剑飞[1] 

机构地区：[1]浙江工业大学计算机软件学院,杭州310013 [2]浙江大学计算机科学与技术学院,杭州310027

出　　处：《计算机辅助设计与图形学学报》2014年第7期1033-1040,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：浙江省自然科学基金(LY12F02037;Y1090335);浙江省科技厅公益技术开发项目(2012C23122;2013C33055)

摘　　要：现有的人脸素描通常需要已受过专业训练的画师来绘制才能达到自然且逼真的效果,不利于在公共安全管理、影视游戏娱乐等应用的推广.为了更方便地获得高质量的人脸素描,提出一种利用粗略人脸线条获取逼真素描的方法.首先从一般人勾绘的人脸粗略线条图像中提取边缘重叠的图像分块;然后利用局部坐标编码学习线条画图像与素描画图像的空间特征并建立各自的数据字典,联合2个数据字典获取线条与其对应素描图像的空间特征关系;进而利用图像分块构造马尔科夫随机场,应用模拟退火算法优化全局能量进行素描合成.在人脸素描XM2VTS数据集上的实验结果表明,采用该方法获得的素描图像效果良好.Existing face sketch usually requires well-trained artist to draw natural and realistic images, which hinders its wide usages in digital entertainment and public security management. We present an approach to automatically generate face sketches based on face rough drawing. Firstly, the input face rough drawing is divided into many overlapped patches. Then, the relation of patches between the rough drawing and the sketch is obtained by a coupled dictionary, which is learned from a set of training data using locality-constrained linear coding. Finally, using simulated annealing algorithm, sketches are synthesized from selected candidate sketch patches and optimized by a global energy optimization based on a Markov random field framework. We evaluate our approach on the face sketch XM2VTS database. Experimental results validate the effectiveness of the proposed method.

关 键 词：图像分块 人脸素描 联合字典 全局能量优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]