本章利用face-recognition库进行图片、视频端的人脸识别
环境:win10,python3.6,dlib19.8.1,face_recognition1.2.2
安装包链接:https://pypi.org/simple/dlib/
下载完成之后,cmd到对应文件夹:
pip install dlib-19.8.1-cp36-cp36m-win_amd64.whl
pip install face_recognition
1.工程下新建facelib文件夹,文件夹中放入目标人物图片,如下图(我放置了四张彭于晏的图片):
内容为:图片名 + 人名,如下图:
(完整代码工程包含单张图片识别与摄像头实时识别,链接:https://download.csdn.net/download/qq_34374211/10588501)
import numpy as np
import cv2
import time
import face_recognition
# Threshold = 0.65 # 人脸置信度阈值
'''
功能:计算两张图片的相似度,范围:[0,1]
输入:
1)人脸A的特征向量
2)人脸B的特征向量
输出:
1)sim:AB的相似度
'''
def simcos(A,B):
A=np.array(A)
B=np.array(B)
dist = np.linalg.norm(A - B) # 二范数
sim = 1.0 / (1.0 + dist) #
return sim
'''
功能:
输入:
1)x:人脸库向量(n维)
2)y:被测人脸的特征向量(1维)
输出:
1)match:与人脸库匹配列表,如[False,True,True,False]
表示被测人脸y与人脸库x的第2,3张图片匹配,与1,4不匹配
2)max(ressim):最大相似度
'''
def compare_faces(x,y,Threshold):
ressim = []
match = [False]*len(x)
for fet in x:
sim = simcos(fet,y)
ressim.append(sim)
if max(ressim) > Threshold: #置信度阈值
match[ressim.index(max(ressim))] = True
return match,max(ressim)
'''
注册身份
输入:
1)libpath:人脸库地址
输出:
1)known_face_encodings:人脸库特征向量
2)known_face_names:人脸库名字标签
'''
def registeredIdentity(libpath):
known_face_encodings, known_face_names = [], []
with open(libpath + 'liblist.txt', 'r') as f:
lines = f.readlines()
for line in lines:
img_lable_name = line.split()
image = face_recognition.load_image_file(libpath + str(img_lable_name[0]))
face_locations = face_recognition.face_locations(image)
# face_locations = face_recognition.face_locations(image, model='cnn')
face_encoding = face_recognition.face_encodings(image, face_locations)[0]
# face_encoding = face_recognition.face_encodings(image, face_locations)
known_face_encodings.append(face_encoding)
known_face_names.append(str(img_lable_name[1]))
return known_face_encodings, known_face_names
'''
输入:
1)testimg:测试图片
2)known_face_encodings:人脸库特征向量
3)known_face_names:人脸库名字标签
输出:
1)retname:预测的名字
2)retscore:相似度得分
3)face_locations:人脸位置坐标
'''
def identityRecognition(testimg,known_face_encodings,known_face_names,Threshold):
face_locations = face_recognition.face_locations(testimg)
# face_locations = face_recognition.face_locations(testimg, model="cnn")
face_encodings = face_recognition.face_encodings(testimg, face_locations)
retname, retscore = "Noface", 0
for face_encoding in face_encodings:
matches, score = compare_faces(known_face_encodings, face_encoding,Threshold)
retname, retscore = "Unknow", 0
if True in matches:
first_match_index = matches.index(True)
name = known_face_names[first_match_index]
if score > retscore:
retname = name
retscore = score
return retname, retscore,face_locations
'''
输入:
1)img:摄像头得到的未裁剪图片
2)face_locations:人脸位置坐标
3) name:预测的名字
输出:
img:加框加年龄备注之后的画面
'''
def age_show(img , face_locations,name):
for (y0, x1, y1, x0) in face_locations:
cv2.rectangle(img, (x0, y0), (x1, y1), ( 0, 0,255), 2)
info = str(name)
t_size = cv2.getTextSize(str(info), cv2.FONT_HERSHEY_PLAIN, 1, 2)[0]
x2,y2 = x0 + t_size[0] + 3, y0 + t_size[1] + 4
cv2.rectangle(img, (x0,y0), (x2,y2), (0, 0, 255), -1) # -1填充作为文字框底色
cv2.putText(img, info, (x0, y0 +t_size[1]+4), cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 1)
return img
'''
功能:识别image身份
输入:
1)image:被测图片
2)libpath:人脸库地址
3)save_dir:图片保存地址
4)Threshold:人脸相似度阈值,Threshold越高识别越精准,但是检出率越低
'''
def pic(image,libpath,save_dir,Threshold=0.65):
known_face_encodings, known_face_names = registeredIdentity(libpath)
retname, retscore,face_locations = identityRecognition(image[:, :, ::-1], known_face_encodings, known_face_names,Threshold=Threshold)
image = age_show(image, face_locations, retname)
print('识别为:%s\n相似度:%2f' % (str(retname), retscore))
cv2.imwrite(save_dir, image)
cv2.imshow('img', image)
cv2.waitKey(0)
调用pic()函数即可识别