現今幾年來，研究人員、學者與業界在於人臉影像的研究上，多著重於偵測和追蹤，對於特徵的擷取亦多應用於電腦辨識，人臉特徵註冊和變形多為娛樂性產品的附加功能。

本論文主要是結合了人臉影像特徵擷取、影像註冊和仿射變形等技術，開發出一個多功能地人臉特徵擷取、影像註冊與變形系統，該系統在現實生活當中許多層面上都蘊含著相當大的發展和應用性。像是可以運用於2D或3D的人像模擬、合成，可以3D模擬人臉術前術後的模樣，又讓夫妻預測下一代的模樣，亦或可以用於史料分析，透過模擬、組合人臉特徵盡可能還原出古人的外型。亦可搭配擴增實境，透過特製裝置投射在使用者身上，讓使用者得到更直接的感受。而近年來全球盛行的美妝與保養風潮、美容、整形等等領域，皆可運用人臉特徵註冊和變形達到更精準地分析、預測和模擬效果。本系統除了常見的全域人臉特徵變換，還開發了13種針對人臉細部特徵的功能，如何讓各項功能在不同人種、膚色等狀態下執行在功能開發上具備相當高的挑戰性。

本研究利用開源機械學習資料庫訓練出來的兩組人臉特徵模組，並搭配自行開發的演算法來進行人臉影像特徵擷取，在配合特徵點進行影像註冊與變形，最後使用無縫融合技術讓擷取特徵能與目標影像完美結合，以此來讓人臉影像的五官、上下眼皮、額頭、眉毛等等特徵可以進行變換或變形。

For the past few years, scientists and companies focus on the research of facial image for detection and trace. Facial image segmentation regularly used for face recognition. Image registration and image transforms usually used on some applications for entertainment.

This study presents a multifunctional system of facial landmark detection, registration, swapping and face morphing. This system which combines the techniques of Human facial image segmentation, Image registration and Affine transformation is full of development and applicability. For example, based on the facial synthesis and simulation system, patients can compare their 2D or 3D portrait before and after the surgery, parents can predict how their child looks like. Furthermore, we can even recover the face of ancient human for historical research. Moreover, patients will directly see and feel the effect of surgery by collaborating the technology of facial image and AR(Augmented Reality) device. In recent years, skin care, face lifting and plastic surgery are prevalent around the world. Our facial landmark detection, registration, swapping and face morphing system can accurately predicts and analyses the effect of medical cosmetology surgery. Despite the common face exchange, we also develop 13 different functions for partial human face. It is a gauntlet to make all the functions work on different skin color or feature.

In this work, we trained two facial landmark modules on open source library and used our algorithm to do image segmentation. Then registered and transformed images by facial landmarks. Finally, we merged source and target image by the technique of image seamless clone.

[1] http://www.chimayclinic.com.tw/cosmeticsurgeryv iew.asp?id=167&sid=308.

[2] https://news.gamme.com.tw/1419384.

[3] Davis E. King Dlib-ml: A Machine Learning Toolkit Journal of Machine Learn- ing Research http://dlib.net/ vol.10:1755-1758, (2009).
[4] Bradski, G. Dr. Dobb’s Journal of Software Tools (2000).

[5] Vahid Kazemi and Josephine Sullivan One Millisecond Face Alignment with an Ensemble of Regression Trees IEEE Conf. Computer Vision and Pattern Recognition,pp.1867-1874(2014).
[6] Xing Wang Swap out a face in one image with a completely different face using OpenCV and DLib. 基 於opencv和Dlib的 人 臉 交 換(face swap) http://blog.csdn.net/wangxing233/article/details/51771265 (2016).

[7] Satya Mallick Face Swap using OpenCV ( C++ / Python )
http://www.learnopencv.com/face-swap-using-opencv-c-python/ (2016).

[8] Patrick Prez, Michel Gangnet, and Andrew Blake. Poisson image editing. In ACM Transactions on Graphics (TOG) vol.22:313–318, ACM, (2003).
[9] H. Lin, P. Du, W. Zhao, L. Zhang, and H. Sun, Image Registration Based on Corner Detection And Affine Transformation. in International Congress on Image and Signal Processing (CISP2010) (Normal University of China, China), pp. 2184-2188, (2010).

[6] Xing Wang Swap out a face in one image with a completely dierent
face using OpenCV and DLib. opencv.Dlib.盡幻(face swap)
http://blog.csdn.net/wangxing233/article/details/51771265 (2016).
[7] Satya Mallick Face Swap using OpenCV ( C++ / Python )
http://www.learnopencv.com/face-swap-using-opencv-c-python/ (2016).
[8] Patrick Prez, Michel Gangnet, and Andrew Blake. Poisson image editing. In
ACM Transactions on Graphics (TOG) vol.22:313318, ACM, (2003).
[9] H. Lin, P. Du, W. Zhao, L. Zhang, and H. Sun, Image Registration Based
on Corner Detection And Ane Transformation. in International Congress on
Image and Signal Processing (CISP2010) (Normal University of China, China),
pp. 2184-2188, (2010).