CV: Face Recognition

Members: Jiaqin Yan, Siyi Wang, Xi Chen

Introduction

We are going to perform a face recogition. The major task is given a face image, find all the face from the same person in image database. Actually it is multi-label classification problem, because we need to treat all the face image from the same person as one label. There are two major challenges:

1. Similarity of faces. Face has features that is easy to distinguish from the background, but distinguish different people's face is hard.

2. Different expressions of faces. Recognising face images from the same person with different expressions is really hard.

Environment setup

Python version: 3.7 Package needed: deepface, keras, numpy

You can use the following commands to setup your environment:

conda create --name CVProj python=2.7
conda install deepface
conda install numpy
conda install keras
conda activate CVProj

File Structure

video.py

Save the video by frame as an image dataset

face_recognization.py

The script includes function for face verification(verifyTest), function for face recognition(findTest),function for face detection(representTest) and funtion for analysis(analysisTest). The script is mainly used to recognize faces in a given dataset.

eval.py

The script can be used to evaluate the best model and the best model metric.

pretrain.py

Using VGG-Face model as start point, fine-tune the model with a part of VGG Face2 dataset