使用dlib人脸检测模型进行人脸活体检测：眨眼+张口
一、眨眼检测
基本原理：计算眼睛长宽比 Eye Aspect Ratio，EAR.当人眼睁开时，EAR在某个值上下波动，当人眼闭合时，EAR迅速下降，理论上会接近于零，当时人脸检测模型还没有这么精确。所以我们认为当EAR低于某个阈值时，眼睛处于闭合状态。为检测眨眼次数，需要设置同一次眨眼的连续帧数。眨眼速度比较快，一般1~3帧就完成了眨眼动作。两个阈值都要根据实际情况设置。
代码如下：
from imutils.video import FileVideoStream
from imutils.video import VideoStream
from imutils import face_utils
import numpy as np
import imutils
import dlib
import cv2
import sys
def _help():
print("Usage:")
print(" python blink_detect.py")
print(" python blink_detect.py ")
print("For example:")
print(" python blink_detect.py video/lee.mp4")
print("If the path of a video is not provided, the camera will be used as the input.Press q to quit.")
def eye_aspect_ratio(eye):
A = np.linalg.norm(eye[1] - eye[5])
B = np.linalg.norm(eye[2] - eye[4])
C = np.linalg.norm(eye[0] - eye[3])
ear = (A + B) / (2.0 * C)
return ear
def blink_detection(vs, file_stream):
# define three constants, one for the eye aspect ratio to indicate
# blink and then the other constants for the min/max number of consecutive
# frames the eye must be below the threshold
EAR_THRESH = 0.2
EAR_CONSEC_FRAMES_MIN = 1
EAR_CONSEC_FRAMES_MAX = 2
# initialize the frame counters and the total number of blinks
blink_counter = [0, 0] # left eye and right eye
blink_total = [0, 0] # left eye and right eye
print("[INFO] loading facial landmark predictor...")
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("model/shape_predictor_68_face_landmarks.dat")
# grab the indexes of the facial landmarks for the left and
# right eye, respectively
(lStart, lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]
(rStart, rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]
print("[INFO] starting video stream thread...")
while True:
# if this is a file video stream, then we need to check if
# there any more frames left in the buffer to process
if file_stream and not vs.more():
break
frame = vs.read()
if frame is not None:
frame = imutils.resize(frame)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
rects = detector(gray, 0)
if len(rects) == 1:
rect = rects[0]
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
left_eye = shape[lStart:lEnd]
right_eye = shape[rStart:rEnd]
left_ear = eye_aspect_ratio(left_eye)
right_ear = eye_aspect_ratio(right_eye)
# compute the convex hull for the left and right eye, then
# visualize each of the eyes
left_eye_hull = cv2.convexHull(left_eye)
right_eye_hull = cv2.convexHull(right_eye)
cv2.drawContours(frame, [left_eye_hull], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [right_eye_hull], -1, (0, 255, 0), 1)
# check to see if the eye aspect ratio is below the blink
# threshold, and if so, increment the blink frame counter
if left_ear < EAR_THRESH:
blink_counter[0] += 1
# otherwise, the eye aspect ratio is not below the blink
# threshold
else:
# if the eyes were closed for a sufficient number of
# then increment the total number of blinks
if EAR_CONSEC_FRAMES_MIN <= blink_counter[0] and blink_counter[0] <= EAR_CONSEC_FRAMES_MAX:
blink_total[0] += 1
blink_counter[0] = 0
# draw the total number of blinks on the frame along with
# the computed eye aspect ratio for the frame
cv2.putText(frame, "LBlinks: {}".format(blink_total[0]), (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "LEAR: {:.2f}".format(left_ear), (10, 60),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
# check to see if the eye aspect ratio is below the blink
# threshold, and if so, increment the blink frame counter
if right_ear < EAR_THRESH:
blink_counter[1] += 1
# otherwise, the eye aspect ratio is not below the blink
# threshold
else:
# if the eyes were closed for a sufficient number of
# then increment the total number of blinks
if EAR_CONSEC_FRAMES_MIN <= blink_counter[1] and blink_counter[1] <= EAR_CONSEC_FRAMES_MAX:
blink_total[1] += 1
blink_counter[1] = 0
# draw the total number of blinks on the frame along with
# the computed eye aspect ratio for the frame
cv2.putText(frame, "RBlinks: {}".format(blink_total[1]), (200, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "REAR: {:.2f}".format(right_ear), (200, 60),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
elif len(rects) == 0:
cv2.putText(frame, "No face!", (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
else:
cv2.putText(frame, "More than one face!", (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.namedWindow("Frame", cv2.WINDOW_NORMAL)
cv2.imshow("Frame", frame)
# if the `q` key was pressed, break from the loop
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()
vs.stop()
if len(sys.argv) > 2 or "-h" in sys.argv or "--help" in sys.argv:
_help()
elif len(sys.argv) == 2:
vs = FileVideoStream(sys.argv[1]).start()
file_stream = True
blink_detection(vs, file_stream)
else:
vs = VideoStream(src=0).start()
file_stream = False
blink_detection(vs, file_stream)
二、张口检测
检测原理：类似眨眼检测，计算Mouth Aspect Ratio,MAR.当MAR大于设定的阈值时，认为张开了嘴巴。
代码如下：
from imutils.video import FileVideoStream
from imutils.video import VideoStream
from imutils import face_utils
import numpy as np
import imutils
import dlib
import cv2
import sys
def _help():
print("Usage:")
print(" python mouth_open_detect.py")
print(" python mouth_open_detect.py ")
print("For example:")
print(" python mouth_open_detect.py video/lee.mp4")
print("If the path of a video is not provided, the camera will be used as the input.Press q to quit.")
def mouth_aspect_ratio(mouth):
A = np.linalg.norm(mouth[2] - mouth[9]) # 51, 59
B = np.linalg.norm(mouth[4] - mouth[7]) # 53, 57
C = np.linalg.norm(mouth[0] - mouth[6]) # 49, 55
mar = (A + B) / (2.0 * C)
return mar
def mouth_open_detection(vs, file_stream):
MAR_THRESH = 0.5
print("[INFO] loading facial landmark predictor...")
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("model/shape_predictor_68_face_landmarks.dat")
(mStart, mEnd) = face_utils.FACIAL_LANDMARKS_IDXS["mouth"]
print("[INFO] starting video stream thread...")
while True:
if file_stream and not vs.more():
break
frame = vs.read()
if frame is not None:
frame = imutils.resize(frame, width=450)
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
rects = detector(gray, 0)
for rect in rects:
shape = predictor(gray, rect)
shape = face_utils.shape_to_np(shape)
mouth = shape[mStart:mEnd]
mar = mouth_aspect_ratio(mouth)
mouth_hull = cv2.convexHull(mouth)
cv2.drawContours(frame, [mouth_hull], -1, (0, 255, 0), 1)
if mar > MAR_THRESH:
cv2.putText(frame, "Mouth is open!", (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "MAR: {:.2f}".format(mar), (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 len(sys.argv) > 2 or "-h" in sys.argv or "--help" in sys.argv:
_help()
elif len(sys.argv) == 2:
vs = FileVideoStream(sys.argv[1]).start()
file_stream = True
mouth_open_detection(vs, file_stream)
else:
vs = VideoStream(src=0).start()
file_stream = False
mouth_open_detection(vs, file_stream)
三、眨眼检测+张口检测
from imutils.video import FileVideoStream
from imutils.video import VideoStream
from imutils import face_utils
import numpy as np
import dlib
import cv2
import sys
def _help():
print("Usage:")
print(" python liveness_detect.py")
print(" python liveness_detect.py ")
print("For example:")
print(" python liveness_detect.py video/lee.mp4")
print("If the path of a video is not provided, the camera will be used as the input.Press q to quit.")
def eye_aspect_ratio(eye):
# (|e1-e5|+|e2-e4|) / (2|e0-e3|)
A = np.linalg.norm(eye[1] - eye[5])
B = np.linalg.norm(eye[2] - eye[4])
C = np.linalg.norm(eye[0] - eye[3])
ear = (A + B) / (2.0 * C)
return ear
def mouth_aspect_ratio(mouth):
# (|m2-m9|+|m4-m7|)/(2|m0-m6|)
A = np.linalg.norm(mouth[2] - mouth[9]) # 51, 59
B = np.linalg.norm(mouth[4] - mouth[7]) # 53, 57
C = np.linalg.norm(mouth[0] - mouth[6]) # 49, 55
mar = (A + B) / (2.0 * C)
return mar
def liveness_detection(vs, file_stream):
EAR_THRESH = 0.15
EAR_CONSEC_FRAMES_MIN = 1
EAR_CONSEC_FRAMES_MAX = 2
MAR_THRESH = 0.5
# 初始化眨眼的连续帧数以及总的眨眼次数
blink_counter = 0
blink_total = 0
print("[INFO] loading facial landmark predictor...")
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("model/shape_predictor_68_face_landmarks.dat")
(lStart, lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]
(rStart, rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]
(mStart, mEnd) = face_utils.FACIAL_LANDMARKS_IDXS["mouth"]
print("[INFO] starting video stream thread...")
while True:
# if this is a file video stream, then we need to check if
# there any more frames left in the buffer to process
if file_stream and not vs.more():
break
frame = vs.read()
if frame is not None:
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
rects = detector(gray, 0)
# 只能处理一张人脸
if len(rects) == 1:
shape = predictor(gray, rects[0]) # 保存68个特征点坐标的对象
shape = face_utils.shape_to_np(shape) # 将shape转换为numpy数组，数组中每个元素为特征点坐标
left_eye = shape[lStart:lEnd]
right_eye = shape[rStart:rEnd]
left_ear = eye_aspect_ratio(left_eye)
right_ear = eye_aspect_ratio(right_eye)
ear = (left_ear + right_ear) / 2.0
mouth = shape[mStart:mEnd]
mar = mouth_aspect_ratio(mouth)
left_eye_hull = cv2.convexHull(left_eye)
right_eye_hull = cv2.convexHull(right_eye)
mouth_hull = cv2.convexHull(mouth)
cv2.drawContours(frame, [left_eye_hull], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [right_eye_hull], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [mouth_hull], -1, (0, 255, 0), 1)
# EAR低于阈值，有可能发生眨眼，眨眼连续帧数加一次
if ear < EAR_THRESH:
blink_counter += 1
# EAR高于阈值，判断前面连续闭眼帧数，如果在合理范围内，说明发生眨眼
else:
# if the eyes were closed for a sufficient number of
# then increment the total number of blinks
if EAR_CONSEC_FRAMES_MIN <= blink_counter and blink_counter <= EAR_CONSEC_FRAMES_MAX:
blink_total += 1
blink_counter = 0
cv2.putText(frame, "Blinks: {}".format(blink_total), (0, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.putText(frame, "Mouth: {}".format("open" if mar > MAR_THRESH else "closed"),
(130, 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.putText(frame, "MAR: {:.2f}".format(mar), (450, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
elif len(rects) == 0:
cv2.putText(frame, "No face!", (0, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
else:
cv2.putText(frame, "More than one face!", (0, 30),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.namedWindow("Frame", cv2.WINDOW_NORMAL)
cv2.imshow("Frame", frame)
# 按下q键退出循环（鼠标要点击一下图片使图片获得焦点）
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()
vs.stop()
if len(sys.argv) > 2 or "-h" in sys.argv or "--help" in sys.argv:
_help()
elif len(sys.argv) == 2:
video_stream = FileVideoStream(sys.argv[1]).start()
file_stream = True
liveness_detection(video_stream, file_stream)
else:
video_stream = VideoStream(src=0).start()
file_stream = False
liveness_detection(video_stream, file_stream) 四、人脸识别破解方法
- 注入应用攻击：在程序中设置断点，通过不断演示人脸识别流程来触发该断点，然后分析并修改程序存储的值，最终使得静态照片也能通过活体检测
- 照片攻击：利用合法用户的照片进行验证
- 视频攻击：利用视频合成软件将合法用户的照片合成为视频
- 3D建模攻击：制作合法用户的脸部3D模型
- 脸部模具攻击
- 利用接口防护不当和设计缺陷
防攻击方式：
- 多重验证
- 识别伪造痕迹
- 提高验证速度
参考博客：
https://www.pyimagesearch.com/2017/04/24/eye-blink-detection-opencv-python-dlib/
https://github.com/mauckc/mouth-open