ZhengRui / FaceAlignmentJNI

A face detection and alignment library on Android

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Example

ex1 ex2

Steps

  1. calibrate according orientation, camera index, check this post.

  2. face detection using opencv haar cascade detector. (scaled to max edge size = 480)

  3. filtering: for each face from previous step, apply skin filter + dlib hog based face detector. dlib detector is slower but with less false positive. (scaled to max edge size = 960)

  4. landmarks detection using dlib, very fast. (original size)

  5. alignment: 68 points template image, their coordinates can be found here. You can either use the template coordinates or your targeted aligned landmarks coordinates (e.g. scale and translation on top of template coordinates) to do affine transformation and cropping. The cropped images will be aligned and saved for feature extraction (if needed).

Setup

  1. modify relevant paths in faceDetAlignLib/build.sh and faceDetAlignLib/CMakeLists.txt, run build.sh

  2. download dlib landmarks detection model from here

  3. adb push shape_predictor_68_face_landmarks.dat /sdcard/FaceDetAlignApp/

Reference

About

A face detection and alignment library on Android


Languages

Language:C++ 49.2%Language:Java 45.6%Language:CMake 3.3%Language:Shell 1.9%