單張無拍攝限制影像重建三維人臉模型

3D Face Model Reconstruction Based on a Single Unconstrained Image

吳梓麟(Zi-Lynn Wu)

指導教授 ： 石勝文

國立暨南國際大學/科技學院/資訊工程學系/碩士(2019年)

摘要

本論文主要實現從單張無拍攝限制人臉影像重建三維人臉模型。基於 Wei et al. 的論文，提出修正過的三維人臉模型重建方法。研究中使用的人臉模型是取自 BP4D-Spontanous 資料庫，此一三維人臉資料庫包含不同人種受測者在不同表情下的高解析度影像與對應的三維模型。三維重建的流程包含攝影機參數估測、參考人臉模型的權重估算、人臉參數最佳化、以及影像貼合與遺失區域補齊等四個主要的步驟。其中除了以 NNLS (Non-Negative Least Squares) 計算人臉模型權重為原有方法，其餘部份均依照問題本身數學特性提出新的方法，以提升數值穩定度及計算速度。在影像貼合與遺失區域補齊階段，加入膚色偵測的調整，能將遺失區域補齊更臻完善，且重建一個人臉模型的平均時間約為 135.40 毫秒。此方法重建出的正臉影像使用人臉辨識系統測試之後，可以使辨識正確率增加 0.02%。

關鍵字

人臉對齊 ； 人臉校正 ； 三維人臉模型 ； 重建三維人臉模型 ； 三維人臉估測 ； 透視投影攝影機校正 ； 非線性最佳化

並列摘要

This thesis aims to reconstruct a 3D face model from a single unconstrained image. A 3D face reconstruction method is developed based on the method proposed by Wei et al. The face models used in this work are selected from the BP4D-Spontanous dataset, which contains high resolution color facial images and the corresponding 3D models of several human races in different facial expressions. The reconstruction method consists of camera parameters estimation, 3D reference model coefficients estimation, nonlinear optimization, texture mapping, and synthesis of occluded image regions. A new method to improve numerical stability in estimating camera parameters is proposed. The 3D reference model coefficients are re-parametrized so as to convert the original constrained optimization problem into an unconstrained one. Also, a skin-color-aware occluded region synthesis method is developed to reduce artifacts in the reconstructed texture map. The proposed method only cost 135.40 milliseconds to reconstruct a 3D face model from an input 2D facial image. Experiments show that the reconstructed 3D face model can be used to improve the face recognition accuracy for 0.02%.

並列關鍵字

Face Alignment ； 3D Face Model ； 3D Face Model Reconstruction ； 3D Face Estimation ； Perspective Projection ； Camera Calibration ； Nonlinear Optimization

參考文獻

[1] X. Zhang, L. Yin, J. F. Cohn, S. Canavan, M. Reale, A. Horowitz, and P. Liu, “A high-resolution spontaneous 3d dynamic facial expression database,” in Automatic Face and Gesture Recognition (FG), 2013 10th IEEE International Conference and Workshops on. IEEE, 2013, pp. 1–6.

[2] X. Zhang, L. Yin, J. F. Cohn, S. Canavan, M. Reale, A. Horowitz, P. Liu, and J. M. Girard, “Bp4d-spontaneous: a high-resolution spontaneous 3d dynamic facial expression database,” Image and Vision Computing, vol. 32, no. 10, pp. 692–706, 2014.

[3] G. B. Huang, M. Ramesh, T. Berg, and E. Learned-Miller, “Labeled Faces in the Wild: ADatabaseforStudyingFaceRecognitioninUnconstrainedEnvironments,” University of Massachusetts, Amherst, Tech. Rep. 07-49, October 2007.

[4] G. B. Huang and E. Learned-Miller, “Labeled Faces in the Wild: Updates and New Reporting Procedures,” University of Massachusetts, Amherst, Tech. Rep. UM-CS-2014-003, May 2014.

[5] C.-P. Wei and Y.-C. F. Wang, “With one look: 3D face shape estimation from a single snapshot,” in 2016 IEEE International Conference on Multimedia and Expo (ICME). IEEE, 2016, pp. 1–6.