Table of contents
1.2 Prepare the required documents and pictures
2. Face recognition development
2.1 Enter your own face information
2.2 Extract the entered facial features
2.3 Capture and recognize faces in real time
The following are some of the things I wanted to do on a whim to play with. I downloaded the face recognition code on github (the URL is attached below). After using it, I suddenly wanted to try the photo recognition effect, and found that the photo will also be recognized. As me, I checked relevant information, usually through liveness detection (such as blinking, turning the head, etc.), and wanted to realize real-time video detection through algorithms to distinguish real people from photos.
1. Preparation
Development environment: windows10+pycharm2022.1.3+python+3.9
1.1 Required libraries
opencv-python==4.5.5.64
numpy==1.23.3
dlib==19.24.0
pandas=1.5.0
pillow=9.2.0
dlib library installation method:
After entering the virtual environment, enter:
anaconda search -t conda dlib
conda install -c https://conda.anaconda.org/conda-forge dlib
1.2 Prepare the required documents and pictures
Enter the URL: http://dlib.net/files/
Download the file: shape_predictor_68_face_landmarks.dat.bz2 , there is a data folder after decompression, the model can recognize 68 key points of the face. The website also contains 5 key point model files.
Prepare face photos and save them in the img folder.
1.3 Test procedure
# -*- coding = utf-8 -*-
# @Time : 2022/9/20 9:11
# @Author : 李昊芸
# @file : dlib_t.py
# @Software : PyCharm
import cv2
import dlib
path = "img/lhy_1.jpeg"
img = cv2.imread(path)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#人脸分类器
detector = dlib.get_frontal_face_detector()
# 获取人脸检测器
predictor = dlib.shape_predictor(
".\\data\\shape_predictor_68_face_landmarks.dat"
)
dets = detector(gray, 1)
for face in dets:
shape = predictor(img, face) # 寻找人脸的68个标定点
# 遍历所有点，打印出其坐标，并圈出来
for pt in shape.parts():
pt_pos = (pt.x, pt.y)
cv2.circle(img, pt_pos, 2, (0, 255, 0), 1)
cv2.imshow("image", img)
cv2.waitKey(0)
cv2.destroyAllWindows()68 key point classifier recognition results:
Multi-person recognition effect:
2. Face recognition development
Original github address: https://github.com/coneypo/Dlib_face_recognition_from_camera
2.1 Enter your own face information
get_faces_from_camera.py
# Copyright (C) 2018-2021 coneypo
# SPDX-License-Identifier: MIT
# Author: coneypo
# Blog: http://www.cnblogs.com/AdaminXie
# GitHub: https://github.com/coneypo/Dlib_face_recognition_from_camera
# Mail: [email protected]
# 进行人脸录入 / Face register
import dlib
import numpy as np
import cv2
import os
import shutil
import time
import logging
# Dlib 正向人脸检测器 / Use frontal face detector of Dlib
detector = dlib.get_frontal_face_detector()
class Face_Register:
def __init__(self):
self.path_photos_from_camera = "data/data_faces_from_camera/"
self.font = cv2.FONT_ITALIC
self.existing_faces_cnt = 0 # 已录入的人脸计数器 / cnt for counting saved faces
self.ss_cnt = 0 # 录入 personX 人脸时图片计数器 / cnt for screen shots
self.current_frame_faces_cnt = 0 # 录入人脸计数器 / cnt for counting faces in current frame
self.save_flag = 1 # 之后用来控制是否保存图像的 flag / The flag to control if save
self.press_n_flag = 0 # 之后用来检查是否先按 'n' 再按 's' / The flag to check if press 'n' before 's'
# FPS
self.frame_time = 0
self.frame_start_time = 0
self.fps = 0
self.fps_show = 0
self.start_time = time.time()
# 新建保存人脸图像文件和数据 CSV 文件夹 / Mkdir for saving photos and csv
def pre_work_mkdir(self):
# 新建文件夹 / Create folders to save face images and csv
if os.path.isdir(self.path_photos_from_camera):
pass
else:
os.mkdir(self.path_photos_from_camera)
# 删除之前存的人脸数据文件夹 / Delete old face folders
def pre_work_del_old_face_folders(self):
# 删除之前存的人脸数据文件夹, 删除 "/data_faces_from_camera/person_x/"...
folders_rd = os.listdir(self.path_photos_from_camera)
for i in range(len(folders_rd)):
shutil.rmtree(self.path_photos_from_camera+folders_rd[i])
if os.path.isfile("data/features_all.csv"):
os.remove("data/features_all.csv")
# 如果有之前录入的人脸, 在之前 person_x 的序号按照 person_x+1 开始录入 / Start from person_x+1
def check_existing_faces_cnt(self):
if os.listdir("data/data_faces_from_camera/"):
# 获取已录入的最后一个人脸序号 / Get the order of latest person
person_list = os.listdir("data/data_faces_from_camera/")
person_num_list = []
for person in person_list:
person_num_list.append(int(person.split('_')[-1]))
self.existing_faces_cnt = max(person_num_list)
# 如果第一次存储或者没有之前录入的人脸, 按照 person_1 开始录入 / Start from person_1
else:
self.existing_faces_cnt = 0
# 更新 FPS / Update FPS of Video stream
def update_fps(self):
now = time.time()
# 每秒刷新 fps / Refresh fps per second
if str(self.start_time).split(".")[0] != str(now).split(".")[0]:
self.fps_show = self.fps
self.start_time = now
self.frame_time = now - self.frame_start_time
self.fps = 1.0 / self.frame_time
self.frame_start_time = now
# 生成的 cv2 window 上面添加说明文字 / PutText on cv2 window
def draw_note(self, img_rd):
# 添加说明 / Add some notes
cv2.putText(img_rd, "Face Register", (20, 40), self.font, 1, (255, 255, 255), 1, cv2.LINE_AA)
cv2.putText(img_rd, "FPS: " + str(self.fps_show.__round__(2)), (20, 100), self.font, 0.8, (0, 255, 0), 1,
cv2.LINE_AA)
cv2.putText(img_rd, "Faces: " + str(self.current_frame_faces_cnt), (20, 140), self.font, 0.8, (0, 255, 0), 1, cv2.LINE_AA)
cv2.putText(img_rd, "N: Create face folder", (20, 350), self.font, 0.8, (255, 255, 255), 1, cv2.LINE_AA)
cv2.putText(img_rd, "S: Save current face", (20, 400), self.font, 0.8, (255, 255, 255), 1, cv2.LINE_AA)
cv2.putText(img_rd, "Q: Quit", (20, 450), self.font, 0.8, (255, 255, 255), 1, cv2.LINE_AA)
# 获取人脸 / Main process of face detection and saving
def process(self, stream):
# 1. 新建储存人脸图像文件目录 / Create folders to save photos
self.pre_work_mkdir()
# 2. 删除 "/data/data_faces_from_camera" 中已有人脸图像文件
# / Uncomment if want to delete the saved faces and start from person_1
# if os.path.isdir(self.path_photos_from_camera):
# self.pre_work_del_old_face_folders()
# 3. 检查 "/data/data_faces_from_camera" 中已有人脸文件
self.check_existing_faces_cnt()
while stream.isOpened():
flag, img_rd = stream.read() # Get camera video stream
kk = cv2.waitKey(1)
faces = detector(img_rd, 0) # Use Dlib face detector
# 4. 按下 'n' 新建存储人脸的文件夹 / Press 'n' to create the folders for saving faces
if kk == ord('n'):
self.existing_faces_cnt += 1
current_face_dir = self.path_photos_from_camera + "person_" + str(self.existing_faces_cnt)
os.makedirs(current_face_dir)
logging.info("\n%-40s %s", "新建的人脸文件夹 / Create folders:", current_face_dir)
self.ss_cnt = 0 # 将人脸计数器清零 / Clear the cnt of screen shots
self.press_n_flag = 1 # 已经按下 'n' / Pressed 'n' already
# 5. 检测到人脸 / Face detected
if len(faces) != 0:
# 矩形框 / Show the ROI of faces
for k, d in enumerate(faces):
# 计算矩形框大小 / Compute the size of rectangle box
height = (d.bottom() - d.top())
width = (d.right() - d.left())
hh = int(height/2)
ww = int(width/2)
# 6. 判断人脸矩形框是否超出 480x640 / If the size of ROI > 480x640
if (d.right()+ww) > 640 or (d.bottom()+hh > 480) or (d.left()-ww < 0) or (d.top()-hh < 0):
cv2.putText(img_rd, "OUT OF RANGE", (20, 300), self.font, 0.8, (0, 0, 255), 1, cv2.LINE_AA)
color_rectangle = (0, 0, 255)
save_flag = 0
if kk == ord('s'):
logging.warning("请调整位置 / Please adjust your position")
else:
color_rectangle = (255, 255, 255)
save_flag = 1
cv2.rectangle(img_rd,
tuple([d.left() - ww, d.top() - hh]),
tuple([d.right() + ww, d.bottom() + hh]),
color_rectangle, 2)
# 7. 根据人脸大小生成空的图像 / Create blank image according to the size of face detected
img_blank = np.zeros((int(height*2), width*2, 3), np.uint8)
if save_flag:
# 8. 按下 's' 保存摄像头中的人脸到本地 / Press 's' to save faces into local images
if kk == ord('s'):
# 检查有没有先按'n'新建文件夹 / Check if you have pressed 'n'
if self.press_n_flag:
self.ss_cnt += 1
for ii in range(height*2):
for jj in range(width*2):
img_blank[ii][jj] = img_rd[d.top()-hh + ii][d.left()-ww + jj]
cv2.imwrite(current_face_dir + "/img_face_" + str(self.ss_cnt) + ".jpg", img_blank)
logging.info("%-40s %s/img_face_%s.jpg", "写入本地 / Save into：",
str(current_face_dir), str(self.ss_cnt))
else:
logging.warning("请先按 'N' 来建文件夹, 按 'S' / Please press 'N' and press 'S'")
self.current_frame_faces_cnt = len(faces)
# 9. 生成的窗口添加说明文字 / Add note on cv2 window
self.draw_note(img_rd)
# 10. 按下 'q' 键退出 / Press 'q' to exit
if kk == ord('q'):
break
# 11. Update FPS
self.update_fps()
cv2.namedWindow("camera", 1)
cv2.imshow("camera", img_rd)
def run(self):
# cap = cv2.VideoCapture("video.mp4") # Get video stream from video file
cap = cv2.VideoCapture(0) # Get video stream from camera
self.process(cap)
cap.release()
cv2.destroyAllWindows()
def main():
logging.basicConfig(level=logging.INFO)
Face_Register_con = Face_Register()
Face_Register_con.run()
if __name__ == '__main__':
main()2.2 Extract the entered facial features
features_extraction_to_csv.py
# Copyright (C) 2018-2021 coneypo
# SPDX-License-Identifier: MIT
# Author: coneypo
# Blog: http://www.cnblogs.com/AdaminXie
# GitHub: https://github.com/coneypo/Dlib_face_recognition_from_camera
# Mail: [email protected]
# 从人脸图像文件中提取人脸特征存入 "features_all.csv" / Extract features from images and save into "features_all.csv"
import os
import dlib
import csv
import numpy as np
import logging
import cv2
# 要读取人脸图像文件的路径 / Path of cropped faces
path_images_from_camera = "data/data_faces_from_camera/"
# Dlib 正向人脸检测器 / Use frontal face detector of Dlib
detector = dlib.get_frontal_face_detector()
# Dlib 人脸 landmark 特征点检测器 / Get face landmarks
predictor = dlib.shape_predictor('data/data_dlib/shape_predictor_68_face_landmarks.dat')
# Dlib Resnet 人脸识别模型，提取 128D 的特征矢量 / Use Dlib resnet50 model to get 128D face descriptor
face_reco_model = dlib.face_recognition_model_v1("data/data_dlib/dlib_face_recognition_resnet_model_v1.dat")
# 返回单张图像的 128D 特征 / Return 128D features for single image
# Input: path_img <class 'str'>
# Output: face_descriptor <class 'dlib.vector'>
def return_128d_features(path_img):
img_rd = cv2.imread(path_img)
faces = detector(img_rd, 1)
logging.info("%-40s %-20s", "检测到人脸的图像 / Image with faces detected:", path_img)
# 因为有可能截下来的人脸再去检测，检测不出来人脸了, 所以要确保是 检测到人脸的人脸图像拿去算特征
# For photos of faces saved, we need to make sure that we can detect faces from the cropped images
if len(faces) != 0:
shape = predictor(img_rd, faces[0])
face_descriptor = face_reco_model.compute_face_descriptor(img_rd, shape)
else:
face_descriptor = 0
logging.warning("no face")
return face_descriptor
# 返回 personX 的 128D 特征均值 / Return the mean value of 128D face descriptor for person X
# Input: path_face_personX <class 'str'>
# Output: features_mean_personX <class 'numpy.ndarray'>
def return_features_mean_personX(path_face_personX):
features_list_personX = []
photos_list = os.listdir(path_face_personX)
if photos_list:
for i in range(len(photos_list)):
# 调用 return_128d_features() 得到 128D 特征 / Get 128D features for single image of personX
logging.info("%-40s %-20s", "正在读的人脸图像 / Reading image:", path_face_personX + "/" + photos_list[i])
features_128d = return_128d_features(path_face_personX + "/" + photos_list[i])
# 遇到没有检测出人脸的图片跳过 / Jump if no face detected from image
if features_128d == 0:
i += 1
else:
features_list_personX.append(features_128d)
else:
logging.warning("文件夹内图像文件为空 / Warning: No images in%s/", path_face_personX)
# 计算 128D 特征的均值 / Compute the mean
# personX 的 N 张图像 x 128D -> 1 x 128D
if features_list_personX:
features_mean_personX = np.array(features_list_personX, dtype=object).mean(axis=0)
else:
features_mean_personX = np.zeros(128, dtype=object, order='C')
return features_mean_personX
def main():
logging.basicConfig(level=logging.INFO)
# 获取已录入的最后一个人脸序号 / Get the order of latest person
person_list = os.listdir("data/data_faces_from_camera/")
person_list.sort()
with open("data/features_all.csv", "w", newline="") as csvfile:
writer = csv.writer(csvfile)
for person in person_list:
# Get the mean/average features of face/personX, it will be a list with a length of 128D
logging.info("%sperson_%s", path_images_from_camera, person)
features_mean_personX = return_features_mean_personX(path_images_from_camera + person)
if len(person.split('_', 2)) == 2:
# "person_x"
person_name = person
else:
# "person_x_tom"
person_name = person.split('_', 2)[-1]
features_mean_personX = np.insert(features_mean_personX, 0, person_name, axis=0)
# features_mean_personX will be 129D, person name + 128 features
writer.writerow(features_mean_personX)
logging.info('\n')
logging.info("所有录入人脸数据存入 / Save all the features of faces registered into: data/features_all.csv")
if __name__ == '__main__':
main()2.3 Capture and recognize faces in real time
face_reco_from_camera.py
# Copyright (C) 2018-2021 coneypo
# SPDX-License-Identifier: MIT
# Author: coneypo
# Blog: http://www.cnblogs.com/AdaminXie
# GitHub: https://github.com/coneypo/Dlib_face_recognition_from_camera
# Mail: [email protected]
# 摄像头实时人脸识别 / Real-time face detection and recognition
import dlib
import numpy as np
import cv2
import pandas as pd
import os
import time
import logging
from PIL import Image, ImageDraw, ImageFont
# Dlib 正向人脸检测器 / Use frontal face detector of Dlib
detector = dlib.get_frontal_face_detector()
# Dlib 人脸 landmark 特征点检测器 / Get face landmarks
predictor = dlib.shape_predictor('data/data_dlib/shape_predictor_68_face_landmarks.dat')
# Dlib Resnet 人脸识别模型，提取 128D 的特征矢量 / Use Dlib resnet50 model to get 128D face descriptor
face_reco_model = dlib.face_recognition_model_v1("data/data_dlib/dlib_face_recognition_resnet_model_v1.dat")
class Face_Recognizer:
def __init__(self):
self.face_feature_known_list = [] # 用来存放所有录入人脸特征的数组 / Save the features of faces in database
self.face_name_known_list = [] # 存储录入人脸名字 / Save the name of faces in database
self.current_frame_face_cnt = 0 # 存储当前摄像头中捕获到的人脸数 / Counter for faces in current frame
self.current_frame_face_feature_list = [] # 存储当前摄像头中捕获到的人脸特征 / Features of faces in current frame
self.current_frame_face_name_list = [] # 存储当前摄像头中捕获到的所有人脸的名字 / Names of faces in current frame
self.current_frame_face_name_position_list = [] # 存储当前摄像头中捕获到的所有人脸的名字坐标 / Positions of faces in current frame
# Update FPS
self.fps = 0 # FPS of current frame
self.fps_show = 0 # FPS per second
self.frame_start_time = 0
self.frame_cnt = 0
self.start_time = time.time()
self.font = cv2.FONT_ITALIC
self.font_chinese = ImageFont.truetype("simsun.ttc", 30)
# 从 "features_all.csv" 读取录入人脸特征 / Read known faces from "features_all.csv"
def get_face_database(self):
if os.path.exists("data/features_all.csv"):
path_features_known_csv = "data/features_all.csv"
csv_rd = pd.read_csv(path_features_known_csv, header=None)
for i in range(csv_rd.shape[0]):
features_someone_arr = []
self.face_name_known_list.append(csv_rd.iloc[i][0])
for j in range(1, 129):
if csv_rd.iloc[i][j] == '':
features_someone_arr.append('0')
else:
features_someone_arr.append(csv_rd.iloc[i][j])
self.face_feature_known_list.append(features_someone_arr)
logging.info("Faces in Database：%d", len(self.face_feature_known_list))
return 1
else:
logging.warning("'features_all.csv' not found!")
logging.warning("Please run 'get_faces_from_camera.py' "
"and 'features_extraction_to_csv.py' before 'face_reco_from_camera.py'")
return 0
# 计算两个128D向量间的欧式距离 / Compute the e-distance between two 128D features
@staticmethod
def return_euclidean_distance(feature_1, feature_2):
feature_1 = np.array(feature_1)
feature_2 = np.array(feature_2)
dist = np.sqrt(np.sum(np.square(feature_1 - feature_2)))
return dist
# 更新 FPS / Update FPS of Video stream
def update_fps(self):
now = time.time()
# 每秒刷新 fps / Refresh fps per second
if str(self.start_time).split(".")[0] != str(now).split(".")[0]:
self.fps_show = self.fps
self.start_time = now
self.frame_time = now - self.frame_start_time
self.fps = 1.0 / self.frame_time
self.frame_start_time = now
# 生成的 cv2 window 上面添加说明文字 / PutText on cv2 window
def draw_note(self, img_rd):
cv2.putText(img_rd, "Face Recognizer", (20, 40), self.font, 1, (255, 255, 255), 1, cv2.LINE_AA)
cv2.putText(img_rd, "Frame: " + str(self.frame_cnt), (20, 100), self.font, 0.8, (0, 255, 0), 1,
cv2.LINE_AA)
cv2.putText(img_rd, "FPS: " + str(self.fps_show.__round__(2)), (20, 130), self.font, 0.8, (0, 255, 0), 1,
cv2.LINE_AA)
cv2.putText(img_rd, "Faces: " + str(self.current_frame_face_cnt), (20, 160), self.font, 0.8, (0, 255, 0), 1,
cv2.LINE_AA)
cv2.putText(img_rd, "Q: Quit", (20, 450), self.font, 0.8, (255, 255, 255), 1, cv2.LINE_AA)
def draw_name(self, img_rd):
# 在人脸框下面写人脸名字 / Write names under rectangle
img = Image.fromarray(cv2.cvtColor(img_rd, cv2.COLOR_BGR2RGB))
draw = ImageDraw.Draw(img)
for i in range(self.current_frame_face_cnt):
# cv2.putText(img_rd, self.current_frame_face_name_list[i], self.current_frame_face_name_position_list[i], self.font, 0.8, (0, 255, 255), 1, cv2.LINE_AA)
draw.text(xy=self.current_frame_face_name_position_list[i], text=self.current_frame_face_name_list[i], font=self.font_chinese,
fill=(255, 255, 0))
img_rd = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
return img_rd
# 修改显示人名 / Show names in chinese
def show_chinese_name(self):
# Default known name: person_1, person_2, person_3
if self.current_frame_face_cnt >= 1:
# 修改录入的人脸姓名 / Modify names in face_name_known_list to chinese name
self.face_name_known_list[0] = '张三'.encode('utf-8').decode()
# self.face_name_known_list[1] = '张四'.encode('utf-8').decode()
# 处理获取的视频流，进行人脸识别 / Face detection and recognition from input video stream
def process(self, stream):
# 1. 读取存放所有人脸特征的 csv / Read known faces from "features.all.csv"
if self.get_face_database():
while stream.isOpened():
self.frame_cnt += 1
logging.debug("Frame %d starts", self.frame_cnt)
flag, img_rd = stream.read()
faces = detector(img_rd, 0)
kk = cv2.waitKey(1)
# 按下 q 键退出 / Press 'q' to quit
if kk == ord('q'):
break
else:
self.draw_note(img_rd)
self.current_frame_face_feature_list = []
self.current_frame_face_cnt = 0
self.current_frame_face_name_position_list = []
self.current_frame_face_name_list = []
# 2. 检测到人脸 / Face detected in current frame
if len(faces) != 0:
# 3. 获取当前捕获到的图像的所有人脸的特征 / Compute the face descriptors for faces in current frame
for i in range(len(faces)):
shape = predictor(img_rd, faces[i])
self.current_frame_face_feature_list.append(face_reco_model.compute_face_descriptor(img_rd, shape))
# 4. 遍历捕获到的图像中所有的人脸 / Traversal all the faces in the database
for k in range(len(faces)):
logging.debug("For face %d in camera:", k+1)
# 先默认所有人不认识，是 unknown / Set the default names of faces with "unknown"
self.current_frame_face_name_list.append("unknown")
# 每个捕获人脸的名字坐标 / Positions of faces captured
self.current_frame_face_name_position_list.append(tuple(
[faces[k].left(), int(faces[k].bottom() + (faces[k].bottom() - faces[k].top()) / 4)]))
# 5. 对于某张人脸，遍历所有存储的人脸特征
# For every faces detected, compare the faces in the database
current_frame_e_distance_list = []
for i in range(len(self.face_feature_known_list)):
# 如果 person_X 数据不为空
if str(self.face_feature_known_list[i][0]) != '0.0':
e_distance_tmp = self.return_euclidean_distance(self.current_frame_face_feature_list[k],
self.face_feature_known_list[i])
logging.debug(" With person %s, the e-distance is %f", str(i + 1), e_distance_tmp)
current_frame_e_distance_list.append(e_distance_tmp)
else:
# 空数据 person_X
current_frame_e_distance_list.append(999999999)
# 6. 寻找出最小的欧式距离匹配 / Find the one with minimum e-distance
similar_person_num = current_frame_e_distance_list.index(min(current_frame_e_distance_list))
logging.debug("Minimum e-distance with %s: %f", self.face_name_known_list[similar_person_num], min(current_frame_e_distance_list))
if min(current_frame_e_distance_list) < 0.4:
self.current_frame_face_name_list[k] = self.face_name_known_list[similar_person_num]
logging.debug("Face recognition result: %s", self.face_name_known_list[similar_person_num])
else:
logging.debug("Face recognition result: Unknown person")
logging.debug("\n")
# 矩形框 / Draw rectangle
for kk, d in enumerate(faces):
# 绘制矩形框
cv2.rectangle(img_rd, tuple([d.left(), d.top()]), tuple([d.right(), d.bottom()]),
(255, 255, 255), 2)
self.current_frame_face_cnt = len(faces)
# 7. 在这里更改显示的人名 / Modify name if needed
# self.show_chinese_name()
# 8. 写名字 / Draw name
img_with_name = self.draw_name(img_rd)
else:
img_with_name = img_rd
logging.debug("Faces in camera now: %s", self.current_frame_face_name_list)
cv2.imshow("camera", img_with_name)
# 9. 更新 FPS / Update stream FPS
self.update_fps()
logging.debug("Frame ends\n\n")
# OpenCV 调用摄像头并进行 process
def run(self):
# cap = cv2.VideoCapture("video.mp4") # Get video stream from video file
cap = cv2.VideoCapture(0) # Get video stream from camera
cap.set(3, 480) # 640x480
self.process(cap)
cap.release()
cv2.destroyAllWindows()
def main():
# logging.basicConfig(level=logging.DEBUG) # Set log level to 'logging.DEBUG' to print debug info of every frame
logging.basicConfig(level=logging.INFO)
Face_Recognizer_con = Face_Recognizer()
Face_Recognizer_con.run()
if __name__ == '__main__':
main()
Recognition results (me and my photos without beauty):
(Mosaic your own face)
It can be seen that the recognition effect is still very good.
3. Liveness detection
3.1 Blink Detection
After querying the data, a widely used indicator is obtained: Eye Aspect Ratio (EAR)
The aspect ratios of the two eyes were respectively obtained and averaged as an indicator of blinking. After several tests, 0.3 was selected as the threshold.
When the EAR is detected twice in a row that is less than the threshold, that is, when the eyes are opened and then closed, we will record it as a blink.
The code is as follows, and the experimental results are shown in the figure:
from scipy.spatial import distance as dist
from imutils.video import VideoStream
from imutils import face_utils
import imutils
import time
import dlib
import cv2
def EAR(eye):
# 计算眼睛的两组垂直关键点之间的欧式距离
A = dist.euclidean(eye[1], eye[5]) # 1,5是一组垂直关键点
B = dist.euclidean(eye[2], eye[4]) # 2,4是一组
# 计算眼睛的一组水平关键点之间的欧式距离
C = dist.euclidean(eye[0], eye[3]) # 0,3是一组水平关键点
return (A + B) / (2.0 * C)
def main():
EAR_THRESH = 0.3 # 眨眼阈值
EYE_close = 2 # 闭眼次数阈值
# 初始化眨眼帧计数器和总眨眼次数
count_eye = 0
total = 0
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
# 左右眼的索引
(lStart, lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]
(rStart, rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]
vs = VideoStream(src=0).start()
time.sleep(1.0)
while True:
frame = vs.read()
frame = imutils.resize(frame, width=600)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 在灰度框中检测人脸
rects = detector(gray, 0)
# 进入循环
for rect in rects:
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
# 提取左眼和右眼坐标，然后使用该坐标计算两只眼睛的眼睛纵横比
leftEye = shape[lStart:lEnd]
rightEye = shape[rStart:rEnd]
ear = EAR(leftEye) + EAR(rightEye) / 2.0
# 判断眼睛纵横比是否低于眨眼阈值
if ear < EAR_THRESH:
count_eye += 1
else:
# 检测到一次闭眼
if count_eye >= EYE_close:
total += 1
count_eye = 0
# 画出画框上眨眼的总次数以及计算出的帧的眼睛纵横比
cv2.putText(frame, "Blinks: {}".format(total), (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "EAR: {:.2f}".format(ear), (300, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
key = cv2.waitKey(1) & 0xFF
if key == ord("q"):
break
cv2.destroyAllWindows()
vs.stop()
if __name__ == '__main__':
main()
3.2 Mouth Opening Detection
In the same way, mouth opening detection can be done:
from imutils.video import VideoStream
from imutils import face_utils
import imutils
import time
import dlib
import cv2
import numpy as np
def MAR(mouth):
# 默认二范数：求特征值，然后求最大特征值得算术平方根
A = np.linalg.norm(mouth[2] - mouth[9]) # 51, 59（人脸68个关键点）
B = np.linalg.norm(mouth[4] - mouth[7]) # 53, 57
C = np.linalg.norm(mouth[0] - mouth[6]) # 49, 55
return (A + B) / (2.0 * C)
def main():
MAR_THRESH = 0.5 # 张嘴阈值
# 初始化
COUNTER_MOUTH = 0
TOTAL_MOUTH = 0
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
# 嘴的索引
(mStart, mEnd) = face_utils.FACIAL_LANDMARKS_IDXS["mouth"]
vs = VideoStream(src=0).start()
time.sleep(1.0)
while True:
frame = vs.read()
frame = imutils.resize(frame, width=600)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 在灰度框中检测人脸
rects = detector(gray, 0)
# 进入循环
for rect in rects:
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
# 提取嘴唇坐标，然后使用该坐标计算嘴唇纵横比
Mouth = shape[mStart:mEnd]
mar = MAR(Mouth)
# 判断嘴唇纵横比是否高于张嘴阈值，如果是，则增加张嘴帧计数器
if mar > MAR_THRESH:
COUNTER_MOUTH += 1
else:
# 如果张嘴帧计数器不等于0，则增加张嘴的总次数
if COUNTER_MOUTH >= 2:
TOTAL_MOUTH += 1
COUNTER_MOUTH = 0
cv2.putText(frame, "Mouth is open: {}".format(TOTAL_MOUTH), (10, 60),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "MAR: {:.2f}".format(mar), (300, 60),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
key = cv2.waitKey(1) & 0xFF
if key == ord("q"):
break
cv2.destroyAllWindows()
vs.stop()
if __name__ == '__main__':
main()3.3 Head shake detection
In many mobile apps such as Alipay, shaking the head is also a very common means of liveness detection. This method uses the change of the Euclidean distance from the nose to the left and right faces to judge whether the head is shaking.
code show as below:
from scipy.spatial import distance as dist
from imutils.video import VideoStream
from imutils import face_utils
import imutils
import time
import dlib
import cv2
def nose_jaw_distance(nose, jaw):
# 计算鼻子上一点"27"到左右脸边界的欧式距离
face_left1 = dist.euclidean(nose[0], jaw[0]) # 27, 0
face_right1 = dist.euclidean(nose[0], jaw[16]) # 27, 16
# 计算鼻子上一点"30"到左右脸边界的欧式距离
face_left2 = dist.euclidean(nose[3], jaw[2]) # 30, 2
face_right2 = dist.euclidean(nose[3], jaw[14]) # 30, 14
# 创建元组，用以保存4个欧式距离值
face_distance = (face_left1, face_right1, face_left2, face_right2)
return face_distance
def main():
# 初始化眨眼帧计数器和总眨眼次数
distance_left = 0
distance_right = 0
TOTAL_FACE = 0
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
(nStart, nEnd) = face_utils.FACIAL_LANDMARKS_IDXS["nose"]
(jStart, jEnd) = face_utils.FACIAL_LANDMARKS_IDXS['jaw']
vs = VideoStream(src=0).start()
time.sleep(1.0)
while True:
frame = vs.read()
frame = imutils.resize(frame, width=600)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 在灰度框中检测人脸
rects = detector(gray, 0)
# 进入循环
for rect in rects:
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
# 提取鼻子和下巴的坐标，然后使用该坐标计算鼻子到左右脸边界的欧式距离
nose = shape[nStart:nEnd]
jaw = shape[jStart:jEnd]
NOSE_JAW_Distance = nose_jaw_distance(nose, jaw)
# 移植鼻子到左右脸边界的欧式距离
face_left1 = NOSE_JAW_Distance[0]
face_right1 = NOSE_JAW_Distance[1]
face_left2 = NOSE_JAW_Distance[2]
face_right2 = NOSE_JAW_Distance[3]
# 根据鼻子到左右脸边界的欧式距离，判断是否摇头
# 左脸大于右脸
if face_left1 >= face_right1 + 2 and face_left2 >= face_right2 + 2:
distance_left += 1
# 右脸大于左脸
if face_right1 >= face_left1 + 2 and face_right2 >= face_left2 + 2:
distance_right += 1
# 左脸大于右脸，并且右脸大于左脸，判定摇头
if distance_left != 0 and distance_right != 0:
TOTAL_FACE += 1
distance_right = 0
distance_left = 0
# 画出摇头次数
cv2.putText(frame, "shake one's head: {}".format(TOTAL_FACE), (10, 90),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
key = cv2.waitKey(1) & 0xFF
if key == ord("q"):
break
cv2.destroyAllWindows()
vs.stop()
if __name__ == '__main__':
main()