MagicMorpher

• A tool for face morphing
• Algorithm is based on this website
• Basically, we firstly do face feature detection and then do Delaunay triangulation. Finally we blend corresponding traingular in two photos to target picture.
• Code is commented in detail

Result Showing

Initial Photo (Hillary and Trump)

<img src="./pic/clinton.jpg" alt="Hillary" width=200px/> <img src="./pic/trump.jpg" alt="Trump" width=200px/>

Feature Detection

<img src="./pic/特征检测.png" alt="Feature Detection" width=400px/>

Delaunay triangulation

<img src="./pic/三角剖分.png" alt="Delaunay triangulation" width=400px/>

Morphing Result

morphing result： <br /> <img src="./pic/morphed_face.png" alt="morph结果" width=400px/>

morphing without boundary <br /> <img src="./pic/no_boundary_morphed_face.png" alt="没有boundary的morph效果" width=400px/>

Usage

./MagicMorpher /path/shape_predictor_68_face_landmarks.dat /path/1.jpg /path/2.jpg /path/target.jpg 0.5 1 0

• First parameter is a pretrained facial landmark detector, you can download from http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2
• Second parameter is the first photo to do morphing
• Third parameter is the second photo to do morphing
• Fourth parameter is target picture's file name
• Fifth parameter alpha is to which degree it agrees with first photo (between 0 and 1, with 1 most liking first photo)
• Sixth parameter is whether to add boundary，(0 or 1)
• Seventh parameter is whether in DemoMode，0 or 1. If 1, then the program will present middle result of feature detection and Delaunay triangulation, Otherwise not.

Code

Dependencies

• opencv3.2.0
• dlib
• libjpeg
• If you want to compile locally, please modify header search paths, framework search paths and library search paths to adjust local opencv, dlib, libjpeg path

Code structure

• FaceLandmarkGetter is to subtract face landmarks
• Trianglationer is to do Delaunay triangulation using face landmarks
• Finally in main.cpp we accomplish mapping to target image

Shortcoming

• Not enough face landmark to do whole head reconstruction
• Lack of feature detection outside of face
• Need both photo to be front look and similar size