Age and Gender Detection Using OpenCV in Python

In this article, we will discuss the process of creating an Age and Gender Predictor using OpenCV. 

Let’s divide the task into 2 parts: 

• Age prediction – The prediction will be in the form of categories where categories are a few age intervals like  0-6,18-25, etc.
• Gender prediction – The prediction is a classifier based where categories are Male and Female.

The approach followed for each task is as follows 

The main problems in this project are the quality of the camera, brightness in the room, the background of the person, the boy cut in the case of girls/the long hair of boys.

The solution to this problem is quite simple, instead of predicting an exact image, we will work on the face of the person only. The flowchart below shows the flow of code.

Now that we have clarity of our problem, let’s go ahead and start with the code.

Import Libraries

• Opencv: To Read and process the image which includes resizing, making a face blob, forming a bounding box, etc.
• Matplotlib: To plot the image.

Python3

import cv2
import matplotlib.pyplot as plt

Importing Image 

To import the image we will be using OpenCV.

Python3

# Input image
image = cv2.imread('image.jpg')
image = cv2.resize(image, (720, 640))

Importing Models 

To download the below pre-trained models follow this link.

Python3

# Importing Models and set mean values
face1 = "opencv_face_detector.pbtxt"
face2 = "opencv_face_detector_uint8.pb"
age1 = "age_deploy.prototxt"
age2 = "age_net.caffemodel"
gen1 = "gender_deploy.prototxt"
gen2 = "gender_net.caffemodel"

MODEL_MEAN_VALUES = (78.4263377603, 87.7689143744, 114.895847746)

# Using models
# Face
face = cv2.dnn.readNet(face2, face1)

# age
age = cv2.dnn.readNet(age2, age1)

# gender
gen = cv2.dnn.readNet(gen2, gen1)

Defining the categories of age and gender in the list.  

Python3

# Categories of distribution
la = ['(0-2)', '(4-6)', '(8-12)', '(15-20)',
      '(25-32)', '(38-43)', '(48-53)', '(60-100)']
lg = ['Male', 'Female']

Copying the image so that we can further use it for demonstrating the final output. 

Python3

# Copy image
fr_cv = image.copy()

Now as we are all set. Let’s start identifying the face blob from the image.  

Python3

# Face detection
fr_h = fr_cv.shape[0]
fr_w = fr_cv.shape[1]
blob = cv2.dnn.blobFromImage(fr_cv, 1.0, (300, 300),
                             [104, 117, 123], True, False)

face.setInput(blob)
detections = face.forward()

Create the bounding box so that further use it in the output image. (For better understanding)

Python3

# Face bounding box creation
faceBoxes = []
for i in range(detections.shape[2]):

    #Bounding box creation if confidence > 0.7
    confidence = detections[0, 0, i, 2]
    if confidence > 0.7:

        x1 = int(detections[0, 0, i, 3]*fr_w)
        y1 = int(detections[0, 0, i, 4]*fr_h)
        x2 = int(detections[0, 0, i, 5]*fr_w)
        y2 = int(detections[0, 0, i, 6]*fr_h)

        faceBoxes.append([x1, y1, x2, y2])

        cv2.rectangle(fr_cv, (x1, y1), (x2, y2),
                      (0, 255, 0), int(round(fr_h/150)), 8)

faceBoxes

Output : 

[[482, 40, 604, 262]]

Finally, implementing the gender and age detection on the face extracted.

Python3

# Checking if face detected or not
if not faceBoxes:
    print("No face detected")

# Final results (otherwise)
# Loop for all the faces detected
for faceBox in faceBoxes:

    #Extracting face as per the faceBox
    face = fr_cv[max(0, faceBox[1]-15):
                 min(faceBox[3]+15, fr_cv.shape[0]-1),
                 max(0, faceBox[0]-15):min(faceBox[2]+15,
                               fr_cv.shape[1]-1)]

    #Extracting the main blob part
    blob = cv2.dnn.blobFromImage(
        face, 1.0, (227, 227), MODEL_MEAN_VALUES, swapRB=False)

    #Prediction of gender
    gen.setInput(blob)
    genderPreds = gen.forward()
    gender = lg[genderPreds[0].argmax()]

    #Prediction of age
    age.setInput(blob)
    agePreds = age.forward()
    age = la[agePreds[0].argmax()]

    #Putting text of age and gender
    #At the top of box
    cv2.putText(fr_cv,
                f'{gender}, {age}',
                (faceBox[0]-150, faceBox[1]+10),
                cv2.FONT_HERSHEY_SIMPLEX,
                1.3,
                (217, 0, 0),
                4,
                cv2.LINE_AA)

    plt.figure(figsize=(7, 7))
    plt.imshow(fr_cv)

Output : 

Output Image

More improvement: 

We can use another pre-trained model like YOLOv5.