There are multiple methods for face detection and face landmark localization. Multi-Task Cascaded Convolutional Neural Network or MTCNN in short is one such method and is based on the paper titled “Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks.” (https://arxiv.org/abs/1604.02878)
The network uses a cascade structure with three networks; first the image is rescaled to a range of different sizes (called an image pyramid), then the first model (Proposal Network or P-Net) proposes candidate facial regions, the second model (Refine Network or R-Net) filters the bounding boxes, and the third model (Output Network or O-Net) proposes facial landmarks
The model is called a multi-task network because each of the three models in the cascade (P-Net, R-Net and O-Net) are trained on three tasks, e.g. make three types of predictions; they are: face classification, bounding box regression, and facial landmark localization.
Open source implementations of the architecture that can be trained on new datasets, as well as pre-trained models that can be used directly for face detection.
-
Download and install python : https://www.python.org/downloads/ Make sure to add python to path
-
Install MTCNN : Open command prompt and use below command
pip install mtcnn -
Tensorflow will be installed along with mtcnn.
If there are any issues with tensorflow uninstall and reinstall tensorflow
pip uninstall tensorflow
pip install tensorflow==2.0
Once the model is configured and loaded, it can be used directly to detect faces in photographs.
Model returns a list of dict objects, each providing a number of keys for the details of each face detected, including:
- 'box' : Providing the x, y of the bottom left of the bounding box, as well as the width and height of the box.
- ‘confidence‘: The probability confidence of the prediction.
- ‘keypoints‘: Providing a dict with dots for the ‘left_eye‘, ‘right_eye‘, ‘nose‘, ‘mouth_left‘, and ‘mouth_right‘.
# Import required headers
import matplotlib.pyplot as plt
from mtcnn.mtcnn import MTCNN
from matplotlib.patches import Rectangle
from matplotlib.patches import Circle
import copyUsing TensorFlow backend.
# draw an image with detected objects
def draw_boxes(filename, result_list):
# load the image
data = plt.imread(filename)
# plot the image
plt.imshow(data)
# get the context for drawing boxes
ax = plt.gca()
# plot each box
for result in result_list:
# get coordinates
x, y, width, height = result['box']
# create the shape
rect = Rectangle((x, y), width, height, fill=False, color='red')
# draw the box
ax.add_patch(rect)
# draw the dots on eyes nose ..
for key, value in result['keypoints'].items():
# create and draw dot
dot = Circle(value, radius=2, color='red')
ax.add_patch(dot)
plt.axis('off')
# show the plot
plt.show()# draw each face separately
def draw_faces(filename, result_list):
# load the image
data1 = plt.imread(filename)
data = copy.deepcopy(data1)
ax = plt.gca()
#Setting up subplots
faceCount = len(result_list)
if(faceCount%4 == 0):
r = faceCount/4
else:
r = faceCount/4 + 1
c = 4
# plot each face as a subplot
for i in range(faceCount):
# get coordinates
x1, y1, width, height = result_list[i]['box']
x2, y2 = x1 + width, y1 + height
# define subplot
#plt.subplot(1, len(result_list), i+1)
plt.subplot(r,c,i+1)
plt.axis('off')
for key, value in result_list[i]['keypoints'].items():
if key=='left_eye':
lx=value[0]
ly=value[1]
if(key=='right_eye'):
rx=value[0]
ry=value[1]
data[ly-2:ly+2,lx-2:lx+2,0]=255
data[ry-2:ry+2,rx-2:rx+2,0]=255
# plot face
plt.imshow(data[y1:y2, x1:x2])
# show the plot
plt.show()#Create the detector(load model), using default weights
detector = MTCNN()#load image from file
filename = "a1.jpeg"
pixels = plt.imread(filename)#detect faces in the image
faces = detector.detect_faces(pixels)#Remove false faces
for face in faces:
confidence = face['confidence']
if(confidence < 0.9):
faces.remove(face)# Draw boxes on faces
draw_boxes(filename, faces)
# Remove individual faces
draw_faces(filename, faces)
#load image from file
filename = "a2.jpeg"
pixels = plt.imread(filename)
#detect faces in the image
faces = detector.detect_faces(pixels)
#Remove false faces
for face in faces:
confidence = face['confidence']
if(confidence < 0.9):
faces.remove(face)# Draw boxes on faces
draw_boxes(filename, faces)
# Remove individual faces
draw_faces(filename, faces)
#load image from file
filename = "a3.jpeg"
pixels = plt.imread(filename)
#detect faces in the image
faces = detector.detect_faces(pixels)
#Remove false faces
for face in faces:
confidence = face['confidence']
if(confidence < 0.9):
faces.remove(face)# Draw boxes on faces
draw_boxes(filename, faces)
# Remove individual faces
draw_faces(filename, faces)
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent