face_recognition项目是世界上最简洁的人脸识别库,可以使用Python和命令行工具提取、识别、操作人脸。
本项目的人脸识别是基于业内领先的C++开源库 dlib中的深度学习模型,用Labeled Faces in the Wild人脸数据集进行测试,有高达99.38%的准确率。但对小孩和亚洲人脸的识别准确率尚待提升。
Labeled Faces in the Wild是美国麻省大学安姆斯特分校(University of Massachusetts Amherst)制作的人脸数据集,该数据集包含了从网络收集的13,000多张面部图像。
本项目提供了简易的face_recognition命令行工具
人脸识别实际上是对人脸进行编码后再去计算两两人脸的相似度,known_image是已知人脸库的图像,unknown_image是待检测的图像,分别利用face_encodings函数来映射成一个向量,再利用两个向量的内积来衡量相似度,compare_faces函数就是根据阈值确认是否是同一人脸。上述函数都是支持多个人脸计算的。另外compare_faces有个tolerance参数是控制阈值的,tolerance值越低越严格,默认为0.6。
import face_recognition之后直接调用face_recognition.load_image_file()读入图像,参数给文件名字符串,注意:jpg文件要和程序放在同一个文件夹下。输出图像是rgb格式(opencv中是bgr格式)。
import face_recognition
#加载图像文件
image = face_recognition.load_image_file("your_file.jpg")
加载图像文件后直接调用face_recognition.face_locations(image),能定位所有图像中识别出的人脸位置信息,返回值是列表形式,列表中每一行是一张人脸的位置信息,包括[top, right, bottom, left],也可理解为每个人脸是一个tuple存储,分别代表框住人脸的矩形中左上角和右下角的坐标(x1,y1,x2,y2)。可遍历列表打印出每张脸的位置信息,也可以通过位置信息截出识别出的人脸的图像显示出来,代码如下:
#定位所有找到的脸的位置
face_locations = face_recognition.face_locations(image)
# 循环找到的所有人脸
for face_location in face_locations:
# 打印每张脸的位置信息
top, right, bottom, left = face_location
print("A face is located at pixel location Top: {}, Left: {}, Bottom: {}, Right: {}".format(top, left, bottom, right))
# 指定人脸的位置信息,然后显示人脸图片
face_image = image[top:bottom, left:right]
pil_image = Image.fromarray(face_image)
pil_image.show()
加载图像后,调用face_recognition.face_landmarks(image)可识别出人脸关键点信息,包括眼睛、鼻子、嘴巴和下巴等,参数仍是加载的图像image,返回值是包含面部特征字典的列表,列表中每一项对应一张人脸,包括nose_bridge、right_eyebrow、right_eye、chine、left_eyebrow、bottom_lip、nose_tip、top_lip、left_eye几个部分,每个部分包含若干个特征点(x,y),总共有68个特征点。列表长度就是图中识别出的人脸数,可遍历此列表和字典的键值对,打印出所有面部特征点,也可在图像上画出来,代码如下:
from PIL import Image, ImageDraw
face_landmarks_list = face_recognition.face_landmarks(image)
pil_image = Image.fromarray(image)
d = ImageDraw.Draw(pil_image)
for face_landmarks in face_landmarks_list:
#face_landmarks_list中每个元素都包含以下key的字典
#打印此图像中每个面部特征的位置
facial_features = [
'chin',
'left_eyebrow',
'right_eyebrow',
'nose_bridge',
'nose_tip',
'left_eye',
'right_eye',
'top_lip',
'bottom_lip'
]
for facial_feature in facial_features:
print("The {} in this face has the following points: {}".format(facial_feature, face_landmarks[facial_feature]))
#在图像中画出每个人脸特征!
for facial_feature in facial_features:
d.line(face_landmarks[facial_feature], width=5)
pil_image.show()
face_recognition.face_encodings(image)方法可获取每个图像文件中每个面部的面部编码,参数仍是加载的图像image,由于每个图像中可能有多个脸,所以返回的是一个编码列表,后续访问时注意加上索引号或者依次遍历。
由打印可看出每张人脸是一个128维的向量。
face_encodings = face_recognition.face_encodings(image)
for face_encoding in face_encodings:
print("face_encoding len = {} \nencoding:{}\n\n".format(len(face_encoding),face_encoding))
用face_recognition.compare_faces()方法可匹配两个面部特征编码,利用两个向量的内积来衡量相似度,根据阈值确认是否是同一人脸。
位置参数:第一个参数给出一个面部编码列表(很多张脸),第二个参数给出单个面部编码(一张脸),compare_faces方法会将第二个参数的编码与第一个参数中的编码依次匹配,返回值是一个布尔值列表,匹配成功的脸返回True,匹配失败的返回False,顺序与第一个参数中脸的顺序一致。
默认参数:tolerance=0.6,可根据自己需求更改,tolerance越小匹配越严格,例如
matches = face_recognition.compare_faces(known_face_encodings, face_encoding,tolerance=0.39),将阈值改为了0.39,阈值太低容易造成无法成功识别人脸,太高容易造成人脸识别混淆,这个是根据我自己测试后慢慢试出来的一个值。
matche_result = face_recognition.compare_faces(known_faces, face_encoding[0],tolerance=0.39)[0]
from PIL import Image, ImageDraw
import face_recognition
# 将jpg文件加载到numpy 数组中
image = face_recognition.load_image_file("view_Mr_G.jpg")
#查找图像中所有面部的所有面部特征
face_landmarks_list = face_recognition.face_landmarks(image)
print("I found {} face(s) in this photograph.".format(len(face_landmarks_list)))
pil_image = Image.fromarray(image)
d = ImageDraw.Draw(pil_image)
for face_landmarks in face_landmarks_list:
#打印此图像中每个面部特征的位置
facial_features = [
'chin',
'left_eyebrow',
'right_eyebrow',
'nose_bridge',
'nose_tip',
'left_eye',
'right_eye',
'top_lip',
'bottom_lip'
]
for facial_feature in facial_features:
print("The {} in this face has the following points: {}".format(facial_feature, face_landmarks[facial_feature]))
#在图像中画出每个人脸特征!
for facial_feature in facial_features:
d.line(face_landmarks[facial_feature], width=5)
pil_image.show()
from PIL import Image
import face_recognition
# 将jpg文件加载到numpy 数组中
image = face_recognition.load_image_file("view_Mr_G.jpg")
# 使用默认的给予HOG模型查找图像中所有人脸
# 这个方法已经相当准确了,但还是不如CNN模型那么准确,因为没有使用GPU加速
# 另请参见: find_faces_in_picture_cnn.py
face_locations = face_recognition.face_locations(image)
# 使用CNN模型
# face_locations = face_recognition.face_locations(image, number_of_times_to_upsample=0, model="cnn")
# 打印:我从图片中找到了 多少 张人脸
print("I found {} face(s) in this photograph.".format(len(face_locations)))
# 循环找到的所有人脸
for face_location in face_locations:
# 打印每张脸的位置信息
top, right, bottom, left = face_location
print("A face is located at pixel location Top: {}, Left: {}, Bottom: {}, Right: {}".format(top, left, bottom, right))
# 指定人脸的位置信息,然后显示人脸图片
face_image = image[top:bottom, left:right]
pil_image = Image.fromarray(face_image)
pil_image.show()
# 识别人脸鉴定是哪个人
import face_recognition
from PIL import Image, ImageDraw
import cv2
PATH = "/Users/tanyashi/Python/python project/face_recognition/一起同过窗"
VIEW_PIC_NAME="view2.jpg"
UNKNOWN_IMAGE="view_Mr_G.jpg"
#将jpg文件加载到numpy数组中
view_image = face_recognition.load_image_file(VIEW_PIC_NAME)
#要识别的图片
unknown_image = face_recognition.load_image_file(UNKNOWN_IMAGE)
results = []
#获取每个图像文件中每个面部的面部编码
#由于每个图像中可能有多个面,所以返回一个编码列表。
#但是由于我知道每个图像只有一个脸,我只关心每个图像中的第一个编码,所以我取索引0。
view_face_encoding = face_recognition.face_encodings(view_image)[0]
print("view_face_encoding:{}\n\n".format(view_face_encoding))
known_faces = [view_face_encoding]
face_locations = face_recognition.face_locations(unknown_image)
print('got {} face(s) in {}:'.format(len(face_locations), UNKNOWN_IMAGE))
for face_location in face_locations:
top, right, bottom, left = face_location
#print(top, right, bottom, left)
face_image = unknown_image[top:bottom, left:right]
face_encoding = face_recognition.face_encodings(face_image)
if face_encoding:
result = {}
result['face_encoding'] = face_encoding
result['is_view'] = face_recognition.compare_faces(known_faces, face_encoding[0])[0]
result['location'] = face_location
result['face_id'] = face_locations.index(face_location)
results.append(result)
if result['is_view']:
print('face {} is view!!'.format(result['face_id']+1))
#print(top, right, bottom, left)
#print("results :{}".format([i['is_view'] for i in results]))
print("please find out view in this image:")
view_face_location = [i['location'] for i in results if i['is_view']]
if view_face_location:
top, right, bottom, left = view_face_location[0]
#print(top, right, bottom, left)
face_image = unknown_image[top:bottom, left:right]
pil_image = Image.fromarray(face_image)
pil_image.show()
else:
print('view is not in this pic!')
# 摄像头头像识别
import face_recognition
import cv2
video_capture = cv2.VideoCapture(0)
# 本地图像
chenduling_image = face_recognition.load_image_file("chenduling.jpg")
chenduling_face_encoding = face_recognition.face_encodings(chenduling_image)[0]
# 本地图像二
sunyizheng_image = face_recognition.load_image_file("sunyizheng.jpg")
sunyizheng_face_encoding = face_recognition.face_encodings(sunyizheng_image)[0]
# 本地图片三
zhangzetian_image = face_recognition.load_image_file("zhangzetian.jpg")
zhangzetian_face_encoding = face_recognition.face_encodings(zhangzetian_image)[0]
# Create arrays of known face encodings and their names
# 脸部特征数据的集合
known_face_encodings = [
chenduling_face_encoding,
sunyizheng_face_encoding,
zhangzetian_face_encoding
]
# 人物名称的集合
known_face_names = [
"michong",
"sunyizheng",
"chenduling"
]
face_locations = []
face_encodings = []
face_names = []
process_this_frame = True
while True:
# 读取摄像头画面
ret, frame = video_capture.read()
# 改变摄像头图像的大小,图像小,所做的计算就少
small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25)
# opencv的图像是BGR格式的,而我们需要是的RGB格式的,因此需要进行一个转换。
rgb_small_frame = small_frame[:, :, ::-1]
# Only process every other frame of video to save time
if process_this_frame:
# 根据encoding来判断是不是同一个人,是就输出true,不是为flase
face_locations = face_recognition.face_locations(rgb_small_frame)
face_encodings = face_recognition.face_encodings(rgb_small_frame, face_locations)
face_names = []
for face_encoding in face_encodings:
# 默认为unknown
matches = face_recognition.compare_faces(known_face_encodings, face_encoding)
name = "Unknown"
# if match[0]:
# name = "michong"
# If a match was found in known_face_encodings, just use the first one.
if True in matches:
first_match_index = matches.index(True)
name = known_face_names[first_match_index]
face_names.append(name)
process_this_frame = not process_this_frame
# 将捕捉到的人脸显示出来
for (top, right, bottom, left), name in zip(face_locations, face_names):
# Scale back up face locations since the frame we detected in was scaled to 1/4 size
top *= 4
right *= 4
bottom *= 4
left *= 4
# 矩形框
cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)
#加上标签
cv2.rectangle(frame, (left, bottom - 35), (right, bottom), (0, 0, 255), cv2.FILLED)
font = cv2.FONT_HERSHEY_DUPLEX
cv2.putText(frame, name, (left + 6, bottom - 6), font, 1.0, (255, 255, 255), 1)
# Display
cv2.imshow('monitor', frame)
# 按Q退出
if cv2.waitKey(1) & 0xFF == ord('q'):
break
video_capture.release()
cv2.destroyAllWindows()
# import face_recognition
import face_recognition
import cv2
import os
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import dlib
import datetime
import threading # 导入threading模块
import time
from imutils import face_utils
from scipy.spatial import distance
import yagmail
class Recorder:
pass
record_dic = {} # 访问记录字典
unknownjpg = [] # 保存来访的陌生人的照片的列表
flagover = 0 # 用于全局判断是否需要保存来访信息
def sendemail(title, contents, fileslist):
yag = yagmail.SMTP(user="2215863456@qq.com", password="agjnddqxmibydjej", host='smtp.qq.com')
yag.send('zeuskkk@163.com', title, contents, fileslist)
# 双眼上下眼距检测
def eye_aspect_ratio(eye):
# print(eye)
A = distance.euclidean(eye[1], eye[5]) # euclidean是计算欧氏距离
B = distance.euclidean(eye[2], eye[4])
C = distance.euclidean(eye[0], eye[3])
ear = (A + B) / (2.0 * C)
return ear
# 字典转换为字符串
def dicttostr():
strlist = []
listkey = list(sorted(record_dic.keys()))
for item in listkey:
strlist.extend([item + ',' + str(onetime) for onetime in record_dic[item].times])
return strlist
def saveRecorder(name, frame):
global record_dic
global flagover
global unknownjpg
if flagover == 1:
return
try:
rec = record_dic[name]
secondsDiff = (datetime.datetime.now() - rec.times[-1]).total_seconds()
if secondsDiff < 60 * 10: # 如果两次比较的时间在10分钟内将被过滤掉
return
rec.times.append(datetime.datetime.now())
print('更新记录', record_dic, rec.times)
except (KeyError):
newRec = Recorder()
newRec.times = [datetime.datetime.now()]
record_dic[name] = newRec
print('添加记录', record_dic, newRec.times)
if name == '未知人员':
s = str(record_dic[name].times[-1])
print('写入', s[:10] + s[-6:])
filename = s[:10] + s[-6:] + '.jpg'
cv2.imwrite(filename, frame)
unknownjpg.append(filename)
def loop_timer_handler(): # 定时器循环触发函数
print('————————Timer headle!————————', str(datetime.datetime.now()))
global timer2
global flagover
global record_dic
global unknownjpg
flagover = 1
timer2 = threading.Timer(60 * 1, loop_timer_handler) # 创建定时器 每1分钟执行一次
timer2.start()
# 如果mail_content不为空,则发送邮件通知
mail_content = '\n'.join(dicttostr())
if mail_content.strip(): # 如果有新的记录内容就发送邮件
sendemail("来访统计记录", mail_content, unknownjpg)
print('来访登记记录邮件已发送', mail_content)
record_dic.clear()
unknownjpg.clear()
print("清空")
time.sleep(10)
print("重新开始")
flagover = 0
timer2 = threading.Timer(2, loop_timer_handler)
timer2.start()
def load_img(sample_dir):
print('加载已知人员图片...')
for (dirpath, dirnames, filenames) in os.walk(sample_dir): # 一级一级的文件夹递归
print(dirpath, dirnames, filenames)
facelib = []
for filename in filenames:
filename_path = os.sep.join([dirpath, filename])
faceimage = face_recognition.load_image_file(filename_path)
# 但是由于我知道每个图像只有一个脸,我只关心每个图像中的第一个编码,所以我取索引0
face_encoding = face_recognition.face_encodings(faceimage)[0]
facelib.append(face_encoding)
return facelib, filenames
def add_text_cv2(text, left, bottom):
'''
由于cv2不支持中文显示,所以在视频流上显示中文时,需要先将帧数据转换成PIL图像,然后利用PIL来添加中文
:param text: 要在图像上添加的文字
:param left: 添加文字离左边框的距离
:param bottom: 添加文字离下边框距离
:return:
'''
global frame
img_PIL = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) # 转换图片格式
font = ImageFont.truetype('simhei.ttf', 40) # 加载字体
position = (left, bottom) # 指定文字输出位置
draw = ImageDraw.Draw(img_PIL) # 绘制照片
draw.text(position, text, font=font, fill=(255, 255, 255)) # 绘制文字
frame = cv2.cvtColor(np.asarray(img_PIL), cv2.COLOR_RGB2BGR) # 将图片转回OpenCV格式
facelib, facename = load_img('../facelib')
video_capture = cv2.VideoCapture(0) # 获得摄像头
face_locations = [] # 定义列表存放人脸位置
face_encodings = [] # 定义列表存放人脸特征编码
process_this_frame = True # 定义信号量
frame_counter = 0 # 连续帧计数
EYE_AR_THRESH = 0.3 # EAR阈值
EYE_AR_CONSEC_FRAMES = 3 # 当EAR小于阈值时,接连多少帧一定发生眨眼动作
# 对应特征点的序号
RIGHT_EYE_START = 37 - 1
RIGHT_EYE_END = 42 - 1
LEFT_EYE_START = 43 - 1
LEFT_EYE_END = 48 - 1
while True:
ret, frame = video_capture.read() # 捕获一帧图片
small_frame = cv2.resize(frame, (0, 0), fx=0.25, fy=0.25) # 将图片缩小1/4,为人脸识别提速
# opencv的图像格式默认是BGR格式,大家可以想象成像素排列跟RGB格式不一样,所以我们必须做一点调整,将像素点进行反向
rgb_small_frame = small_frame[:, :, ::-1] # 将opencv的BGR格式转为RGB格式
# 找到这一帧图像上所有的人脸位置
face_locations = face_recognition.face_locations(rgb_small_frame)
# 得到面部编码(见资料文章)
face_encodings = face_recognition.face_encodings(rgb_small_frame, face_locations)
# print('人脸特征编码长度', len(face_encodings))
face_names = [] # 定义列表,放置识别出来的人员的名字
for face_encoding in face_encodings: # 循环多张人脸, 这里一般只有一张人脸
matches = face_recognition.compare_faces(facelib, face_encoding, tolerance=0.39)
name = "未知人员" # 定义默认的识别结果为Unknown
# 有true说明图片对比已经成功,取出识别的人员名字并开始进行活体眨眼检测
if True in matches: # 如果识别出来,就将名称取出
# 这里matches是个列表,包含每张已知图片的对比结果,这里应该只有一个为true
# 所以我们要找出true的位置,这个位置就是找到的那个人的图片
first_match_index = matches.index(True)
# 取出这个人的名字
name = facename[first_match_index][:-4] # -4是去掉文件后缀
print('开始活体检测...')
# 在图上加上请眨眼的提示
add_text_cv2('{}请眨眼...'.format(name), 0, 0) # 在图像上添加眨眼提示
detector = dlib.get_frontal_face_detector() # 人脸检测器
# 特征检测算法要做个介绍
predictor = dlib.shape_predictor('../libs/shape_predictor_68_face_landmarks.dat') # 人脸特征点检测器
gray = cv2.cvtColor(rgb_small_frame, cv2.COLOR_BGR2GRAY)
# 人脸检测,1代表把图片像素放大1倍以便能够搜集到更多的照片细节,返回检测到所有人脸区域的数组
# 这个数不能设置过大,否则影响运行速度
rects = detector(gray, 1)
if len(rects) > 0:
print('#' * 20)
shape = predictor(gray, rects[0]) # 检测特征点,rect[0]代表取第一个人脸
# 将所有的脸部坐标点转换为numpy array的格式
# convert the facial landmark (x, y)-coordinates to a NumPy array
points = face_utils.shape_to_np(shape)
leftEye = points[LEFT_EYE_START:LEFT_EYE_END + 1] # 取出左眼对应的特征点
rightEye = points[RIGHT_EYE_START:RIGHT_EYE_END + 1] # 取出右眼对应的特征点
leftEAR = eye_aspect_ratio(leftEye) # 计算左眼EAR
rightEAR = eye_aspect_ratio(rightEye) # 计算右眼EAR
ear = (leftEAR + rightEAR) / 2.0 # 求左右眼EAR的均值
# 如果EAR小于阈值,开始计算连续帧,只有连续帧计数超过EYE_AR_CONSEC_FRAMES时,才会计做一次眨眼
if ear < EYE_AR_THRESH:
frame_counter += 1
print('ear小于阈值了!!!', frame_counter)
if frame_counter >= EYE_AR_CONSEC_FRAMES:
# blink_counter += 1
print('眨眼检测成功,请进入!')
frame_counter = 0
print('活体检测结束')
face_names.append(name) # 保存识别结果
# process_this_frame = not process_this_frame #信号量保护结束(这个语句没什么用)
# 显示结果
for (top, right, bottom, left), name in zip(face_locations, face_names):
top *= 4 # 还原人脸的原始尺寸,原来是1/4,现在放到原大小
right *= 4
bottom *= 4
left *= 4
# frame:要显示的帧,(left, top), (right, bottom) 各方向坐标,(0, 0, 255) 人脸框颜色, 2是线条粗细
cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2) # 标注人脸
add_text_cv2(name, left + 6, bottom - 6)
saveRecorder(name, frame) # 过滤并保存记录
cv2.imshow('security check', frame) # 显示图片检测框
if cv2.waitKey(1) & 0xFF == ord('q'): # 等待键盘输入,当输入q时,整个程序退出
break
video_capture.release() # 释放摄像头资源
time.sleep(2) # 休眠10秒
timer2.cancel() # 关闭定时器线程