Face recognition research
written by wooram kang
-
environment && Dependency about research
OS : ubuntu 16.04 language : python 3.5 package : a. CUDA : 9.0 b. cuDNN : 7.1.4 c. opencv d. tensorflow : 1.9.0 e. dlib : 19.15.0 f. numpy g. sklearn h. keras 2.2.3 ...
-
Dataset for training and testing location : /Dataset_Image_Face
a. LFW http://vis-www.cs.umass.edu/lfw/ Labeled Faces in the Wild
13233 images 5749 people 1680 people with two or more
b. VGG VGGFace2 http://zeus.robots.ox.ac.uk/vgg_face2/ Visual Geometry group
3.31 Million images 9131 people 87 - 850 people with two or more
c. VGG VGGFace1 http://www.robots.ox.ac.uk/~vgg/data/vgg_face/ Visual Geometry group
2622 people
cf. tiny version of VGGFace2
caution. each group has diffenrt kinds of image'size for each. so make sure the sizes of groups to be as same as one's size of those.
mainly, a. LFW & b. VGGFace2 be used and i don't provide those dataset. only links i left -
trial algorithms A by A
ToDo list :
a. detection + recognition
1. Deep Face
2. Open Face
3. VGG __face recognition
4. Openbr
5. facenet
6. face-everthing
b. detection only
1. tiny face
2. MTCNN_Face_detection
c. recognition only
1. Fisherfaces
2. shanren7_ real time face recogntion
3. LBPH Algorithm
4. insightface
d. preprocessing
1. remove light on LAB colour system
2. CLAHE
3. gamma correction
d-prime. preprocessing with autoencoder
1. VAE
2. denosing AE
Done list :
a. FaceNet papers review
b. FaceNet modeling
c. FaceNet testing
d. preprocessing by Image processing algorithm
e. including VAE, lots of autoendoers papers review
f. autoencoders modeling
g. autuencoders testing
Final model :
preprocessing with CV
1-1. remove light on LAB colour system
1-2. CLAHE
1-3. gamma correction
2. affine transform
preprocessing with autoencoder
1. denosing AE
detection and discrimination with a model based FACENET
1. FACENET
postprocessing
1. Face tracking