摘要
提出一种融合人脸特征分析的自适应编辑传播方法来提高人脸图像光照渲染的效能,并以此实现复杂背景下基于单张参考人脸的自动光照迁移。该方法把参考图像的人脸区域与目标图像的背景区域进行融合,利用边缘保持平滑滤波从融合的人脸中提取光照信息。然后,构建一个能随不同人脸区域而自适应变化的编辑传播模型,把提取的光照信息从人脸区域扩散到背景区域,生成光照模板。最后,通过融合光照模板与目标人脸实现光照迁移。在YaleB数据库的定量实验中,平均每张迁移光照人脸有超过85%的像素(归一化到[0,255])与标准光照人脸的像素值差异小于6。与其他方法对比,本文方法获得的光照渲染效果具有更好一致性。结果表明,本文方法扩展了光照迁移的适用范围,具有良好的稳定性,能在具有不同性别和背景等特点的参考人脸与目标人脸中生成自然的光照迁移效果。
An adaptive edit propagation method based on facial priors was proposed to achieve natural relighting effect of a portrait in a complex background using a single reference face.Firstly,the facial region of a reference image and the background region of a target were combined,and an edgepreserving smoothing filter was used to extract the illumination information from the combined image.Then a new edit propagation model adaptively changed with facial parameters was constructed to generate an illumination template by propagating the illumination from the facial region to the background.Finally,the illumination template and the target were multiplied in the luminance channel to achieve the relighting effect.The quantitative experiments in YaleB database show that there are averagely over 85% pixels（normalized to [0,255]）in a relighting effect face,whose intensity differences are less than 6 comparing with the ground true.As compared with other methods,the relighting effects of proposed method are more consistency.The conclusion shows that the proposed method achieves reliable and natural face relighting effect on portraits with different genders and backgrounds.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2015年第5期1450-1457,共8页
Optics and Precision Engineering
基金
国家自然科学基金资助项目(No.61472144
No.61273255)
国家科技支撑计划资助项目(No.2013BAH65F01-2013BAH65F04)
国家教育部博士点基金资助项目(No.20120172110023)
中央高校基本科研业务费专项资金资助项目(SCUT 2013ZG0011)
广东省教育厅科技创新项目(No.2013KJCX0010)
中国博士后科学基金资助项目
关键词
人脸图像
光照迁移
编辑传播
边缘保持平滑
光照模板
face image
face relighting
edit propagation
edge-preserving smoothing
illumination template
