一种侧视图的三维人脸重建方法
PDF下载 (357)蒋 玉,赵杰煜*,陈能仑.一种侧视图的三维人脸重建方法[J].宁波大学学报(理工版),2016,29(3):62-67.DOI:
JIANG Yu,ZHAO Jie-yu*,CHEN Neng-lun.A 3D Face Reconstruction Method Using a Single Side View Image[J].Journal of Ningbo University(Natural Science & Engineering Edition),2016,29(3):62-67.DOI:
| Title: | A 3D Face Reconstruction Method Using a Single Side View Image |
| 作者: | 蒋 玉, 赵杰煜*, 陈能仑 |
| Author(s): | JIANG Yu, ZHAO Jie-yu*, CHEN Neng-lun |
| 关键词: | 三维人脸重建; 侧视图; 从运动中恢复结构; 简化的三维形变模型; 三维模型拟合 |
| Keywords: | 3D face reconstruction; side view; SFM; 3DMM; 3D model fitting |
| 分类号: | TP391.4 |
| 文献标识码: | A |
| 摘要: | 三维形变模型(3D Morphable Model, 3DMM)和从运动中恢复结构(Structure From Motion, SFM)方法被广泛用于三维人脸重建. 基于单视图进行三维人脸重建需要正视图和先验模型, 会受到计算复杂度高、容易陷入局部极小值和易受姿态变化的影响. 本文提出一种针对侧视图的三维人脸重建方法, 首先对侧视图使用改进的三维形变模型, 得到初始的三维人脸正视图及特征点; 然后根据人脸对称性, 得到侧视图对称的视图及对应的面部特征点; 最后用SFM方法将正视图、原始视图和对称视图重建, 得到稀疏三维模型. 并用不同姿态的面部图片对该方法进行了评估, 结果表明该方法比已有的方法对姿态变化更具鲁棒性. |
| Abstract: | The 3D morphable model (3DMM) and the Structure from Motion (SFM) method are widely used in 3D face reconstruction. The popular 3D face reconstruction method based on a single image needs a front view image and a priori model. It usually involves high computation complexity, and can be easily affected by posture changes. It is also susceptible to local minimum in algorithm execution. In this paper we propose a 3D face reconstruction method using a single side view image. First of all, we use the improved three-dimensional deformation model to obtain the initial front view and feature points; Secondly, a symmetrical view and the corresponding facial feature points are calculated according to the facial symmetry. Finally, the sparse 3D model is reconstructed using the SFM method given the front view, initial view and symmetric view. Experiments are conducted with facial images of different postures. The results show that the proposed method is more robust to posture changes than other general methods. |
| 参考文献 /References: | [1] ZHANG X, GAO F. Face recognition across pose: A review[J]. Pattern Recognition, 2009, 42(11):2876-2896. [2] MOEINI A, MOEINI A, FAEZ K. Unrestricted pose-invariant face recognition by sparse dictionary matrix[J]. Image and Vision Computing, 2015, 36(7):9- 22. [3] KEMELMACHER-SHLIZERMAN I, BASRI R. 3D face reconstruction from a single image using a single reference face shape[J]. IEEE Trans Pattern Anal Mach Intell, 2011, 33(2):394-405. [4] NARENDRA P, MUKESH Z. 3D facial model re- construction, expressions synthesis and animation using frontal face image[J]. Signal Image and Video Processing, 2013, 7(5):889-897. [5] LEE S, PARK K, KIM J. A SFM-based 3D face reconstruction method robust to self-occlusion by using a shape conversion matrix[J]. Pattern Recognition, 2011, 44(7):1470-1486. [6] LEE Y J, LEE S J, KANG R P, et al. Single view-based 3D face reconstruction robust to self-occlusion[J]. EURASIP (Advances Signal Process), 2012, 10(7):1-20. [7] BLANZ V, VETTER T. Face recognition based on fitting a 3D morphable model[J]. IEEE Trans Pattern Anal Mach Intell, 2003, 25(9):1063-1074. [8] TAIGMAN Y, YANG M, RANZATO M A, et al. Deep face: Closing the gap to human-level performance in face verification[EB/OL]. [2015-06-10]. http://wenku.baidu.com/ view/f3f0dd1ca0116c175f0e48be.html. [9] MAHDI B G, MALEK B, SERGE C. 3D head pose estimation and tracking using particle filtering and ICP algorithm[C]//FRANCISCO J, ROBERT B. Articulated Motion and Deformable Objects, 2010:224-237. [10] ABDULL A H, SAYEE D, MUTH U, et al. Face recognition with symmetric local graph structure[J]. MedSci, 2014, 14(41):6131-6137. [11] ZHOU E J, FAN H Q, CAO Z M, et al. Extensive facial landmark localization with coarse-to-fine convolutional network cascade[C]//Computer Vision Workshops (ICCVW), 2013 IEEE International Conference, Sydney, 2013:386-391. |
| 备注/Memo: | 收稿日期: 2015?08?20. 宁波大学学报（理工版）网址: http://journallg.nbu.edu.cn/ 基金项目: 国家自然科学基金（61175026）; 国家科技支撑计划项目（2012BAF12B11）; 科技部国际科技合作专项（2013DFG12810）; 浙江省国际科技合作专项（2013C24027）. 第一作者: 蒋玉（1989－）, 女, 湖南邵阳人, 在读硕士研究生, 主要研究方向: 计算机视觉和图像处理. E-mail: mearoyuj@foxmail.com *通信作者: 赵杰煜（1965－）, 男, 浙江宁波人, 教授, 主要研究方向: 随机神经网络. E-mail: zhao_jieyu@nbu.edu.cn 宁波大学学报（理工版）网址：http://journallg.nbu.edu.cn/ |
