Face detection and recognition using AI algorithms (writing in progress)

The purpose of this project to recognize detected faces. First of all, it is a question of detecting the faces (answer the question how to detect a face), and then, in a second step, it is a question of recognizing the detected faces (for instance answer the question who is the person in the image). The applications of face recognition are present in the current life: to unlock mobile phones, or even in China to control citizens and to assign behavioral scores to them…

1. Tools and Python modules

The idea is to consider the similarity between the faces: a small distance if it concerns the same person, and a large one if it concerns two different persons. The aim is to obtain a vector that represents a face in a unique way.

I have used several modules:

• matplotlib.pyplot to load images and plot them
• numpy for manipulating vectors and matrices and computing the distances
• face_recognition to have embeddings of images, this module contains several trained deep learning models giving representative vectors of images
• Open-CV because there is computer vision, and it can also be used to draw bounding boxes

import matplotlib.pyplot as plt
import numpy as np
import cv2
import face_recognition

2. Useful steps and functions

For face recognition, it is essential to find a way to compute the faces encodings. For a given person (or myself), I started with a portrait picture to learn embeddings.

picture1 = plt.imread('myPicture1.jpg')
embedding1 = face_recognition.face_encodings(picture1)

picture2 = plt.imread('myPicture2.jpg')
embedding2 = face_recognition.face_encodings(picture2)

picture1 and picture2 are the embeddings of these images. It is possible to play with their outputs but I didn’t change the default parameters.